VictoriaMetrics/docs/VictoriaLogs/LogsQL.md
Aliaksandr Valialkin a98fb495c6
lib/logstorage: allow specifying _time filter offset without time range
This is useful when builiding graphs on time ranges in the past.
2024-11-08 16:43:54 +01:00

153 KiB
Raw Blame History

weight title menu aliases
5 LogsQL
docs
parent weight
victorialogs 5
/VictoriaLogs/LogsQL.html

LogsQL is a simple yet powerful query language for VictoriaLogs. See examples and tutorial in order to feel the language.

LogsQL provides the following features:

LogsQL tutorial

If you aren't familiar with VictoriaLogs, then start with key concepts docs.

Then follow these docs:

The simplest LogsQL query is just a word, which must be found in the log message. For example, the following query finds all the logs with error word:

error

It is recommended to use vlogscli for querying VictoriaLogs.

If the queried word clashes with LogsQL keywords, then just wrap it into quotes. For example, the following query finds all the log messages with and word:

"and"

It is OK to wrap any word into quotes. For example:

"error"

Moreover, it is possible to wrap phrases containing multiple words in quotes. For example, the following query finds log messages with the error: cannot find file phrase:

"error: cannot find file"

Queries above match logs with any timestamp, e.g. they may return logs from the previous year alongside recently ingested logs.

Usually logs from the previous year aren't so interesting comparing to the recently ingested logs. So it is recommended adding time filter to the query. For example, the following query returns logs with the error word, which were ingested into VictoriaLogs during the last 5 minutes:

error AND _time:5m

This query consists of two filters joined with AND operator:

The AND operator means that the log entry must match both filters in order to be selected.

Typical LogsQL query consists of multiple filters joined with AND operator. It may be tiresome typing and then reading all these AND words. So LogsQL allows omitting AND words. For example, the following query is equivalent to the query above:

_time:5m error

The query returns logs in arbitrary order because sorting of big amounts of logs may require non-trivial amounts of CPU and RAM. The number of logs with error word over the last 5 minutes isn't usually too big (e.g. less than a few millions), so it is OK to sort them with sort pipe. The following query sorts the selected logs by _time field:

_time:5m error | sort by (_time)

It is unlikely you are going to investigate more than a few hundreds of logs returned by the query above. So you can limit the number of returned logs with limit pipe. The following query returns the last 10 logs with the error word over the last 5 minutes:

_time:5m error | sort by (_time) desc | limit 10

By default VictoriaLogs returns all the log fields. If you need only the given set of fields, then add fields pipe to the end of the query. For example, the following query returns only _time, _stream and _msg fields:

error _time:5m | fields _time, _stream, _msg

Suppose the query above selects too many rows because some buggy app pushes invalid error logs to VictoriaLogs. Suppose the app adds buggy_app word to every log line. Then the following query removes all the logs from the buggy app, allowing us paying attention to the real errors:

_time:5m error NOT buggy_app

This query uses NOT operator for removing log lines from the buggy app. The NOT operator is used frequently, so it can be substituted with - or ! char (the ! must be used instead of - in front of = and ~ filters like != and !~). The following query is equivalent to the previous one:

_time:5m error -buggy_app

Suppose another buggy app starts pushing invalid error logs to VictoriaLogs - it adds foobar word to every emitted log line. No problems - just add -foobar to the query in order to remove these buggy logs:

_time:5m error -buggy_app -foobar

This query can be rewritten to more clear query with the OR operator inside parentheses:

_time:5m error -(buggy_app OR foobar)

The parentheses are required here, since otherwise the query won't return the expected results. The query error -buggy_app OR foobar is interpreted as (error AND NOT buggy_app) OR foobar according to priorities for AND, OR and NOT operator. This query returns logs with foobar word, even if do not contain error word or contain buggy_app word. So it is recommended wrapping the needed query parts into explicit parentheses if you are unsure in priority rules. As an additional bonus, explicit parentheses make queries easier to read and maintain.

Queries above assume that the error word is stored in the log message. If this word is stored in other field such as log.level, then add log.level: prefix in front of the error word:

_time:5m log.level:error -(buggy_app OR foobar)

The field name can be wrapped into quotes if it contains special chars or keywords, which may clash with LogsQL syntax. Any word also can be wrapped into quotes. So the following query is equivalent to the previous one:

"_time":"5m" "log.level":"error" -("buggy_app" OR "foobar")

What if the application identifier - such as buggy_app and foobar - is stored in the app field? Correct - just add app: prefix in front of buggy_app and foobar:

_time:5m log.level:error -(app:buggy_app OR app:foobar)

The query can be simplified by moving the app: prefix outside the parentheses:

_time:5m log.level:error -app:(buggy_app OR foobar)

The app field uniquely identifies the application instance if a single instance runs per each unique app. In this case it is recommended associating the app field with log stream fields during data ingestion. This usually improves both compression rate and query performance when querying the needed streams via _stream filter. If the app field is associated with the log stream, then the query above can be rewritten to more performant one:

_time:5m log.level:error {app!~"buggy_app|foobar"}

This query skips scanning for log messages from buggy_app and foobar apps. It inspects only log.level and _stream labels. This significantly reduces disk read IO and CPU time needed for performing the query.

LogsQL also provides functions for statistics calculation over the selected logs. For example, the following query returns the number of logs with the error word for the last 5 minutes:

_time:5m error | stats count() logs_with_error

Finally, it is recommended reading performance tips.

Now you are familiar with LogsQL basics. See LogsQL examples and query syntax if you want to continue learning LogsQL.

Key concepts

Word

LogsQL splits all the log fields into words delimited by non-word chars such as whitespace, parens, punctuation chars, etc. For example, the foo: (bar,"тест")! string is split into foo, bar and тест words. Words can contain arbitrary utf-8 chars. These words are taken into account by full-text search filters such as word filter, phrase filter and prefix filter.

Query syntax

LogsQL query must contain at least a single filter for selecting the matching logs. For example, the following query selects all the logs for the last 5 minutes by using _time filter:

_time:5m

Tip: try * filter, which selects all the logs stored in VictoriaLogs. Do not worry - this doesn't crash VictoriaLogs, even if the query selects trillions of logs. See these docs if you are curious why.

Additionally to filters, LogQL query may contain arbitrary mix of optional actions for processing the selected logs. These actions are delimited by | and are known as pipes. For example, the following query uses stats pipe for returning the number of log messages with the error word for the last 5 minutes:

_time:5m error | stats count() errors

See the list of supported pipes in LogsQL.

Filters

LogsQL supports various filters for searching for log messages (see below). They can be combined into arbitrary complex queries via logical filters.

Filters are applied to _msg field by default. If the filter must be applied to other log field, then its' name followed by the colon must be put in front of the filter. For example, if error word filter must be applied to the log.level field, then use log.level:error query.

Field names and filter args can be put into quotes if they contain special chars, which may clash with LogsQL syntax. LogsQL supports quoting via double quotes ", single quotes ' and backticks:

"some 'field':123":i('some("value")') AND `other"value'`

If doubt, it is recommended quoting field names and filter args.

The list of LogsQL filters:

Time filter

VictoriaLogs scans all the logs per each query if it doesn't contain the filter on _time field. It uses various optimizations in order to accelerate full scan queries without the _time filter, but such queries can be slow if the storage contains large number of logs over long time range. The easiest way to optimize queries is to narrow down the search with the filter on _time field.

For example, the following query returns log messages ingested into VictoriaLogs during the last hour, which contain the error word:

_time:1h AND error

The following formats are supported for _time filter:

  • _time:duration matches logs with timestamps on the time range (now-duration, now], where duration can have these values. Examples:
    • _time:5m - returns logs for the last 5 minutes
    • _time:2.5d15m42.345s - returns logs for the last 2.5 days, 15 minutes and 42.345 seconds
    • _time:1y - returns logs for the last year
  • _time:YYYY-MM-DDZ - matches all the logs for the particular day by UTC. For example, _time:2023-04-25Z matches logs on April 25, 2023 by UTC.
  • _time:YYYY-MMZ - matches all the logs for the particular month by UTC. For example, _time:2023-02Z matches logs on February, 2023 by UTC.
  • _time:YYYYZ - matches all the logs for the particular year by UTC. For example, _time:2023Z matches logs on 2023 by UTC.
  • _time:YYYY-MM-DDTHHZ - matches all the logs for the particular hour by UTC. For example, _time:2023-04-25T22Z matches logs on April 25, 2023 at 22 hour by UTC.
  • _time:YYYY-MM-DDTHH:MMZ - matches all the logs for the particular minute by UTC. For example, _time:2023-04-25T22:45Z matches logs on April 25, 2023 at 22:45 by UTC.
  • _time:YYYY-MM-DDTHH:MM:SSZ - matches all the logs for the particular second by UTC. For example, _time:2023-04-25T22:45:59Z matches logs on April 25, 2023 at 22:45:59 by UTC.
  • _time:[min_time, max_time] - matches logs on the time range [min_time, max_time], including both min_time and max_time. The min_time and max_time can contain any format specified here. For example, _time:[2023-04-01Z, 2023-04-30Z] matches logs for the whole April, 2023 by UTC, e.g. it is equivalent to _time:2023-04Z.
  • _time:[min_time, max_time) - matches logs on the time range [min_time, max_time), not including max_time. The min_time and max_time can contain any format specified here. For example, _time:[2023-02-01Z, 2023-03-01Z) matches logs for the whole February, 2023 by UTC, e.g. it is equivalent to _time:2023-02Z.

It is possible to specify time zone offset for all the absolute time formats by appending +hh:mm or -hh:mm suffix. For example, _time:2023-04-25+05:30 matches all the logs on April 25, 2023 by India time zone, while _time:2023-02-07:00 matches all the logs on February, 2023 by California time zone.

If the timezone offset information is missing, then the local time zone of the host where VictoriaLogs runs is used. For example, _time:2023-10-20 matches all the logs for 2023-10-20 day according to the local time zone of the host where VictoriaLogs runs.

It is possible to specify generic offset for the selected time range by appending offset after the _time filter. Examples:

  • _time:offset 1h matches logs until now-1h.
  • _time:5m offset 1h matches logs on the time range (now-1h5m, now-1h].
  • _time:2023-07Z offset 5h30m matches logs on July, 2023 by UTC with offset 5h30m.
  • _time:[2023-02-01Z, 2023-03-01Z) offset 1w matches logs the week before the time range [2023-02-01Z, 2023-03-01Z) by UTC.

Performance tips:

  • It is recommended specifying the smallest possible time range during the search, since it reduces the amounts of log entries, which need to be scanned during the query. For example, _time:1h is usually faster than _time:5h.

  • While LogsQL supports arbitrary number of _time:... filters at any level of logical filters, it is recommended specifying a single _time filter at the top level of the query.

  • See other performance tips.

See also:

Day range filter

_time:day_range[start, end] filter allows returning logs on the particular start ... end time per every day, where start and end have the format hh:mm. For example, the following query matches logs between 08:00 and 18:00 UTC every day:

_time:day_range[08:00, 18:00)

This query includes 08:00, while 18:00 is excluded, e.g. the last matching time is 17:59:59.999999999. Replace [ with ( in order to exclude the starting time. Replace ) with ] in order to include the ending time. For example, the following query matches logs between 08:00 and 18:00, excluding 08:00:00.000000000 and including 18:00:

_time:day_range(08:00, 18:00]

If the time range must be applied to other than UTC time zone, then add offset <duration>, where <duration> can have any supported duration value. For example, the following query selects logs between 08:00 and 18:00 at +0200 time zone:

_time:day_range[08:00, 18:00) offset 2h

Performance tip: it is recommended specifying regular time filter additionally to day_range filter. For example, the following query selects logs between 08:00 and 20:00 every day for the last week:

_time:1w _time:day_range[08:00, 18:00)

See also:

Week range filter

_time:week_range[start, end] filter allows returning logs on the particular start ... end days per every day, where start and end can have the following values:

  • Sun or Sunday
  • Mon or Monday
  • Tue or Tuesday
  • Wed or Wednesday
  • Thu or Thursday
  • Fri or Friday
  • Sat or Saturday

For example, the following query matches logs between Monday and Friday UTC every day:

_time:week_range[Mon, Fri]

This query includes Monday and Friday. Replace [ with ( in order to exclude the starting day. Replace ] with ) in order to exclude the ending day. For example, the following query matches logs between Sunday and Saturday, excluding Sunday and Saturday (e.g. it is equivalent to the previous query):

_time:week_range(Sun, Sat)

If the day range must be applied to other than UTC time zone, then add offset <duration>, where <duration> can have any supported duration value. For example, the following query selects logs between Monday and Friday at +0200 time zone:

_time:week_range[Mon, Fri] offset 2h

The week_range filter can be combined with day_range filter using logical filters. For example, the following query selects logs between 08:00 and 18:00 every day of the week excluding Sunday and Saturday:

_time:week_range[Mon, Fri] _time:day_range[08:00, 18:00)

Performance tip: it is recommended specifying regular time filter additionally to week_range filter. For example, the following query selects logs between Monday and Friday per every week for the last 4 weeks:

_time:4w _time:week_range[Mon, Fri]

See also:

Stream filter

VictoriaLogs provides an optimized way to select logs, which belong to particular log streams. This can be done via {...} filter, which may contain arbitrary Prometheus-compatible label selector over fields associated with log streams. For example, the following query selects log entries with app field equal to nginx:

{app="nginx"}

This query is equivalent to the following exact filter query, but the upper query usually works much faster:

app:="nginx"

It is allowed to add _stream: prefix in front of {...} filter. The following filter is equivalent to {app="nginx"}:

_stream:{app="nginx"}

Performance tips:

  • It is recommended using the most specific {...} filter matching the smallest number of log streams, which needs to be scanned by the rest of filters in the query.

  • While LogsQL supports arbitrary number of {...} filters at any level of logical filters, it is recommended specifying a single {...} filter at the top level of the query.

  • See other performance tips.

See also:

_stream_id filter

Every log stream in VictoriaLogs is uniquely identified by _stream_id field. The _stream_id:... filter allows quickly selecting all the logs belonging to the particular stream.

For example, the following query selects all the logs, which belong to the log stream with _stream_id equal to 0000007b000001c850d9950ea6196b1a4812081265faa1c7:

_stream_id:0000007b000001c850d9950ea6196b1a4812081265faa1c7

If the log stream contains too many logs, then it is good idea limiting the number of returned logs with time filter. For example, the following query selects logs for the given stream for the last hour:

_time:1h _stream_id:0000007b000001c850d9950ea6196b1a4812081265faa1c7

The _stream_id filter supports specifying multiple _stream_id values via _stream_id:in(...) syntax. For example:

_stream_id:in(0000007b000001c850d9950ea6196b1a4812081265faa1c7, 1230007b456701c850d9950ea6196b1a4812081265fff2a9)

It is also possible specifying subquery inside in(...), which selects the needed _stream_id values. For example, the following query returns logs for log streams containing error word in the _msg field during the last 5 minutes:

_stream_id:in(_time:5m error | fields _stream_id)

See also:

Word filter

The simplest LogsQL query consists of a single word to search in log messages. For example, the following query matches log messages with error word inside them:

error

This query matches the following log messages:

  • error
  • an error happened
  • error: cannot open file

This query doesn't match the following log messages:

  • ERROR, since the filter is case-sensitive by default. Use i(error) for this case. See these docs for details.
  • multiple errors occurred, since the errors word doesn't match error word. Use error* for this case. See these docs for details.

By default the given word is searched in the _msg field. Specify the field name in front of the word and put a colon after it if it must be searched in the given field. For example, the following query returns log entries containing the error word in the log.level field:

log.level:error

Both the field name and the word in the query can contain arbitrary utf-8-encoded chars. For example:

სფერო:τιμή

Both the field name and the word in the query can be put inside quotes if they contain special chars, which may clash with the query syntax. For example, the following query searches for the ip 1.2.3.45 in the field ip:remote:

"ip:remote":"1.2.3.45"

See also:

Phrase filter

Is you need to search for log messages with the specific phrase inside them, then just wrap the phrase in quotes. The phrase can contain any chars, including whitespace, punctuation, parens, etc. They are taken into account during the search. For example, the following query matches log messages with ssh: login fail phrase inside them:

"ssh: login fail"

This query matches the following log messages:

  • ERROR: ssh: login fail for user "foobar"
  • ssh: login fail!

This query doesn't match the following log messages:

  • ssh login fail, since the message misses : char just after the ssh. Use seq("ssh", "login", "fail") query if log messages with the sequence of these words must be found. See these docs for details.
  • login fail: ssh error, since the message doesn't contain the full phrase requested in the query. If you need matching a message with all the words listed in the query, then use ssh AND login AND fail query. See these docs for details.
  • ssh: login failed, since the message ends with failed word instead of fail word. Use "ssh: login fail"* query for this case. See these docs for details.
  • SSH: login fail, since the SSH word is in capital letters. Use i("ssh: login fail") for case-insensitive search. See these docs for details.

If the phrase contains double quotes, then either put \ in front of double quotes or put the phrase inside single quotes. For example, the following filter searches logs with "foo":"bar" phrase:

'"foo":"bar"'

By default the given phrase is searched in the _msg field. Specify the field name in front of the phrase and put a colon after it if it must be searched in the given field. For example, the following query returns log entries containing the cannot open file phrase in the event.original field:

event.original:"cannot open file"

Both the field name and the phrase can contain arbitrary utf-8-encoded chars. For example:

შეტყობინება:"Το αρχείο δεν μπορεί να ανοίξει"

The field name can be put inside quotes if it contains special chars, which may clash with the query syntax. For example, the following query searches for the cannot open file phrase in the field some:message:

"some:message":"cannot open file"

See also:

Prefix filter

If you need to search for log messages with words / phrases containing some prefix, then just add * char to the end of the word / phrase in the query. For example, the following query returns log messages, which contain words with err prefix:

err*

This query matches the following log messages:

  • err: foobar
  • cannot open file: error occurred

This query doesn't match the following log messages:

  • Error: foobar, since the Error word starts with capital letter. Use i(err*) for this case. See these docs for details.
  • fooerror, since the fooerror word doesn't start with err. Use ~"err" for this case. See these docs for details.

Prefix filter can be applied to phrases. For example, the following query matches log messages containing phrases with unexpected fail prefix:

"unexpected fail"*

This query matches the following log messages:

  • unexpected fail: IO error
  • error:unexpected failure

This query doesn't match the following log messages:

  • unexpectedly failed, since the unexpectedly doesn't match unexpected word. Use unexpected* AND fail* for this case. See these docs for details.
  • failed to open file: unexpected EOF, since failed word occurs before the unexpected word. Use unexpected AND fail* for this case. See these docs for details.

If the prefix contains double quotes, then either put \ in front of double quotes or put the prefix inside single quotes. For example, the following filter searches logs with "foo":"bar prefix:

'"foo":"bar'*

By default the prefix filter is applied to the _msg field. Specify the needed field name in front of the prefix filter in order to apply it to the given field. For example, the following query matches log.level field containing any word with the err prefix:

log.level:err*

If the field name contains special chars, which may clash with the query syntax, then it may be put into quotes in the query. For example, the following query matches log:level field containing any word with the err prefix.

"log:level":err*

Performance tips:

See also:

Substring filter

If it is needed to find logs with some substring, then ~"substring" filter can be used. For example, the following query matches log entries, which contain ampl text in the _msg field:

~"ampl"

It matches the following messages:

  • Example message
  • This is a sample

It doesn't match EXAMPLE message, since AMPL substring here is in uppercase. Use ~"(?i)ampl" filter instead. Note that case-insensitive filter may be much slower than case-sensitive one.

Performance tip: prefer using word filter and phrase filter, since substring filter may be quite slow.

See also:

Range comparison filter

LogsQL supports field:>X, field:>=X, field:<X and field:<=X filters, where field is the name of log field and X is numeric value, IPv4 address or a string. For example, the following query returns logs containing numeric values for the response_size field bigger than 10*1024:

response_size:>10KiB

The following query returns logs with user field containing string values smaller than John:

username:<"John"

See also:

Empty value filter

Sometimes it is needed to find log entries without the given log field. This can be performed with log_field:"" syntax. For example, the following query matches log entries without host.hostname field:

host.hostname:""

See also:

Any value filter

Sometimes it is needed to find log entries containing any non-empty value for the given log field. This can be performed with log_field:* syntax. For example, the following query matches log entries with non-empty host.hostname field:

host.hostname:*

See also:

Exact filter

The word filter and phrase filter return log messages, which contain the given word or phrase inside them. The message may contain additional text other than the requested word or phrase. If you need searching for log messages or log fields with the exact value, then use the exact filter. For example, the following query returns log messages with the exact value fatal error: cannot find /foo/bar:

="fatal error: cannot find /foo/bar"

The query doesn't match the following log messages:

  • fatal error: cannot find /foo/bar/baz or some-text fatal error: cannot find /foo/bar, since they contain an additional text other than the specified in the exact filter. Use "fatal error: cannot find /foo/bar" query in this case. See these docs for details.

  • FATAL ERROR: cannot find /foo/bar, since the exact filter is case-sensitive. Use i("fatal error: cannot find /foo/bar") in this case. See these docs for details.

By default the exact filter is applied to the _msg field. Specify the field name in front of the exact filter and put a colon after it if it must be searched in the given field. For example, the following query returns log entries with the exact error value at log.level field:

log.level:="error"

Both the field name and the phrase can contain arbitrary utf-8-encoded chars. For example:

log.დონე:="შეცდომა"

The field name can be put inside quotes if it contains special chars, which may clash with the query syntax. For example, the following query matches the error value in the field log:level:

"log:level":="error"

See also:

Exact prefix filter

Sometimes it is needed to find log messages starting with some prefix. This can be done with the ="prefix"* filter. For example, the following query matches log messages, which start from Processing request prefix:

="Processing request"*

This filter matches the following log messages:

  • Processing request foobar
  • Processing requests from ...

It doesn't match the following log messages:

  • processing request foobar, since the log message starts with lowercase p. Use ="processing request"* OR ="Processing request"* query in this case. See these docs for details.
  • start: Processing request, since the log message doesn't start with Processing request. Use "Processing request" query in this case. See these docs for details.

By default the exact filter is applied to the _msg field. Specify the field name in front of the exact filter and put a colon after it if it must be searched in the given field. For example, the following query returns log entries with log.level field, which starts with err prefix:

log.level:="err"*

Both the field name and the phrase can contain arbitrary utf-8-encoded chars. For example:

log.დონე:="შეცდომა"*

The field name can be put inside quotes if it contains special chars, which may clash with the query syntax. For example, the following query matches log:level values starting with err prefix:

"log:level":="err"*

See also:

Multi-exact filter

Sometimes it is needed to locate log messages with a field containing one of the given values. This can be done with multiple exact filters combined into a single logical filter. For example, the following query matches log messages with log.level field containing either error or fatal exact values:

log.level:(="error" OR ="fatal")

While this solution works OK, LogsQL provides simpler and faster solution for this case - the in() filter.

log.level:in("error", "fatal")

It works very fast for long lists passed to in().

It is possible to pass arbitrary query inside in(...) filter in order to match against the results of this query. The query inside in(...) must end with fields pipe containing a single field name, so VictoriaLogs could fetch results from this field. For example, the following query selects all the logs for the last 5 minutes for users, who visited pages with admin word in the path field during the last day:

_time:5m AND user_id:in(_time:1d AND path:admin | fields user_id)

See also:

Case-insensitive filter

Case-insensitive filter can be applied to any word, phrase or prefix by wrapping the corresponding word filter, phrase filter or prefix filter into i(). For example, the following query returns log messages with error word in any case:

i(error)

The query matches the following log messages:

  • unknown error happened
  • ERROR: cannot read file
  • Error: unknown arg
  • An ErRoR occurred

The query doesn't match the following log messages:

  • FooError, since the FooError word has superfluous prefix Foo. Use ~"(?i)error" for this case. See these docs for details.
  • too many Errors, since the Errors word has superfluous suffix s. Use i(error*) for this case.

By default the i() filter is applied to the _msg field. Specify the needed field name in front of the filter in order to apply it to the given field. For example, the following query matches log.level field containing error word in any case:

log.level:i(error)

If the field name contains special chars, which may clash with the query syntax, then it may be put into quotes in the query. For example, the following query matches log:level field containing error word in any case.

"log:level":i("error")

Performance tips:

See also:

Sequence filter

Sometimes it is needed to find log messages with words or phrases in a particular order. For example, if log messages with error word followed by open file phrase must be found, then the following LogsQL query can be used:

seq("error", "open file")

This query matches some error: cannot open file /foo/bar message, since the open file phrase goes after the error word. The query doesn't match the cannot open file: error message, since the open file phrase is located in front of the error word. If you need matching log messages with both error word and open file phrase, then use error AND "open file" query. See these docs for details.

By default the seq() filter is applied to the _msg field. Specify the needed field name in front of the filter in order to apply it to the given field. For example, the following query matches event.original field containing (error, "open file") sequence:

event.original:seq(error, "open file")

If the field name contains special chars, which may clash with the query syntax, then it may be put into quotes in the query. For example, the following query matches event:original field containing (error, "open file") sequence:

"event:original":seq(error, "open file")

See also:

Regexp filter

LogsQL supports regular expression filter with re2 syntax via ~"regex" syntax. For example, the following query returns all the log messages containing err or warn susbstrings:

~"err|warn"

The query matches the following log messages, which contain either err or warn substrings:

  • error: cannot read data
  • 2 warnings have been raised
  • data transferring finished

The query doesn't match the following log messages:

  • ERROR: cannot open file, since the ERROR word is in uppercase letters. Use ~"(?i)(err|warn)" query for case-insensitive regexp search. See these docs for details. See also case-insensitive filter docs.
  • it is warmer than usual, since it doesn't contain neither err nor warn substrings.

If the regexp contains double quotes, then either put \ in front of double quotes or put the regexp inside single quotes. For example, the following regexp searches logs matching "foo":"(bar|baz)" regexp:

'"foo":"(bar|baz)"'

By default the regexp filter is applied to the _msg field. Specify the needed field name in front of the filter in order to apply it to the given field. For example, the following query matches event.original field containing either err or warn substrings:

event.original:~"err|warn"

If the field name contains special chars, which may clash with the query syntax, then it may be put into quotes in the query. For example, the following query matches event:original field containing either err or warn substrings:

"event:original":~"err|warn"

Performance tips:

  • Prefer combining simple word filter with logical filter instead of using regexp filter. For example, the ~"error|warning" query can be substituted with error OR warning query, which usually works much faster. Note that the ~"error|warning" matches errors as well as warnings words, while error OR warning matches only the specified words. See also multi-exact filter.
  • Prefer moving the regexp filter to the end of the logical filter, so lighter filters are executed first.
  • Prefer using ="some prefix"* instead of ~"^some prefix", since the exact filter works much faster than the regexp filter.
  • See other performance tips.

See also:

Range filter

If you need to filter log message by some field containing only numeric values, then the range() filter can be used. For example, if the request.duration field contains the request duration in seconds, then the following LogsQL query can be used for searching for log entries with request durations exceeding 4.2 seconds:

request.duration:range(4.2, Inf)

This query can be shortened to by using range comparison filter:

request.duration:>4.2

The lower and the upper bounds of the range(lower, upper) are excluded by default. If they must be included, then substitute the corresponding parentheses with square brackets. For example:

  • range[1, 10) includes 1 in the matching range
  • range(1, 10] includes 10 in the matching range
  • range[1, 10] includes 1 and 10 in the matching range

The range boundaries can contain any supported numeric values.

Note that the range() filter doesn't match log fields with non-numeric values alongside numeric values. For example, range(1, 10) doesn't match the request took 4.2 seconds log message, since the 4.2 number is surrounded by other text. Extract the numeric value from the message with parse(_msg, "the request took <request_duration> seconds") transformation and then apply the range() filter pipe to the extracted request_duration field.

Performance tips:

See also:

IPv4 range filter

If you need to filter log message by some field containing only IPv4 addresses such as 1.2.3.4, then the ipv4_range() filter can be used. For example, the following query matches log entries with user.ip address in the range [127.0.0.0 - 127.255.255.255]:

user.ip:ipv4_range(127.0.0.0, 127.255.255.255)

The ipv4_range() accepts also IPv4 subnetworks in CIDR notation. For example, the following query is equivalent to the query above:

user.ip:ipv4_range("127.0.0.0/8")

If you need matching a single IPv4 address, then just put it inside ipv4_range(). For example, the following query matches 1.2.3.4 IP at user.ip field:

user.ip:ipv4_range("1.2.3.4")

Note that the ipv4_range() doesn't match a string with IPv4 address if this string contains other text. For example, ipv4_range("127.0.0.0/24") doesn't match request from 127.0.0.1: done log message, since the 127.0.0.1 ip is surrounded by other text. Extract the IP from the message with parse(_msg, "request from <ip>: done") transformation and then apply the ipv4_range() filter pipe to the extracted ip field.

Hints:

  • If you need searching for log messages containing the given X.Y.Z.Q IPv4 address, then "X.Y.Z.Q" query can be used. See these docs for details.
  • If you need searching for log messages containing at least a single IPv4 address out of the given list, then "ip1" OR "ip2" ... OR "ipN" query can be used. See these docs for details.
  • If you need finding log entries with ip field in multiple ranges, then use ip:(ipv4_range(range1) OR ipv4_range(range2) ... OR ipv4_range(rangeN)) query. See these docs for details.

Performance tips:

See also:

String range filter

If you need to filter log message by some field with string values in some range, then string_range() filter can be used. For example, the following LogsQL query matches log entries with user.name field starting from A and B chars:

user.name:string_range(A, C)

The string_range() includes the lower bound, while excluding the upper bound. This simplifies querying distinct sets of logs. For example, the user.name:string_range(C, E) would match user.name fields, which start from C and D chars.

See also:

Length range filter

If you need to filter log message by its length, then len_range() filter can be used. For example, the following LogsQL query matches log messages with lengths in the range [5, 10] chars:

len_range(5, 10)

This query matches the following log messages, since their length is in the requested range:

  • foobar
  • foo bar

This query doesn't match the following log messages:

  • foo, since it is too short
  • foo bar baz abc, sinc it is too long

It is possible to use inf as the upper bound. For example, the following query matches log messages with the length bigger or equal to 5 chars:

len_range(5, inf)

The range boundaries can be expressed in the following forms:

  • Hexadecimal form. For example, len_range(0xff, 0xABCD).
  • Binary form. Form example, len_range(0b100110, 0b11111101)
  • Integer form with _ delimiters for better readability. For example, len_range(1_000, 2_345_678).

By default the len_range() is applied to the _msg field. Put the field name in front of the len_range() in order to apply the filter to the needed field. For example, the following query matches log entries with the foo field length in the range [10, 20] chars:

foo:len_range(10, 20)

See also:

Logical filter

Simpler LogsQL filters can be combined into more complex filters with the following logical operations:

  • q1 AND q2 - matches common log entries returned by both q1 and q2. Arbitrary number of filters can be combined with AND operation. For example, error AND file AND app matches log messages, which simultaneously contain error, file and app words. The AND operation is frequently used in LogsQL queries, so it is allowed to skip the AND word. For example, error file app is equivalent to error AND file AND app.

  • q1 OR q2 - merges log entries returned by both q1 and q2. Arbitrary number of filters can be combined with OR operation. For example, error OR warning OR info matches log messages, which contain at least one of error, warning or info words.

  • NOT q - returns all the log entries except of those which match q. For example, NOT info returns all the log messages, which do not contain info word. The NOT operation is frequently used in LogsQL queries, so it is allowed substituting NOT with - and ! in queries. For example, -info and !info are equivalent to NOT info. The ! must be used instead of - in front of = and ~ filters like != and !~.

The NOT operation has the highest priority, AND has the middle priority and OR has the lowest priority. The priority order can be changed with parentheses. For example, NOT info OR debug is interpreted as (NOT info) OR debug, so it matches log messages, which do not contain info word, while it also matches messages with debug word (which may contain the info word). This is not what most users expect. In this case the query can be rewritten to NOT (info OR debug), which correctly returns log messages without info and debug words.

LogsQL supports arbitrary complex logical queries with arbitrary mix of AND, OR and NOT operations and parentheses.

By default logical filters apply to the _msg field unless the inner filters explicitly specify the needed log field via field_name:filter syntax. For example, (error OR warn) AND host.hostname:host123 is interpreted as (_msg:error OR _msg:warn) AND host.hostname:host123.

It is possible to specify a single log field for multiple filters with the following syntax:

field_name:(q1 OR q2 OR ... qN)

For example, log.level:error OR log.level:warning OR log.level:info can be substituted with the shorter query: log.level:(error OR warning OR info).

Performance tips:

  • VictoriaLogs executes logical operations from the left to the right, so it is recommended moving the most specific and the fastest filters (such as word filter and phrase filter) to the left, while moving less specific and the slowest filters (such as regexp filter and case-insensitive filter) to the right. For example, if you need to find log messages with the error word, which match some /foo/(bar|baz) regexp, it is better from performance PoV to use the query error ~"/foo/(bar|baz)" instead of ~"/foo/(bar|baz)" error.

    The most specific filter means that it matches the lowest number of log entries comparing to other filters.

  • See other performance tips.

Pipes

Additionally to filters, LogsQL query may contain arbitrary mix of '|'-delimited actions known as pipes. For example, the following query uses stats, sort and limit pipes for returning top 10 log streams with the biggest number of logs during the last 5 minutes:

_time:5m | stats by (_stream) count() per_stream_logs | sort by (per_stream_logs desc) | limit 10

LogsQL supports the following pipes:

block_stats pipe

<q> | block_stats pipe returns the following stats per each block processed by <q>. This pipe is needed mostly for debugging.

The returned per-block stats:

  • field - field name
  • rows - the number of rows at the given `field.
  • type - internal storage type for the given field
  • values_bytes - on-disk size of the data for the given field
  • bloom_bytes - on-disk size of bloom filter data for the given field
  • dict_bytes - on-disk size of the dictionary data for the given field
  • dict_items - the number of unique values in the dictionary for the given field

See also:

blocks_count pipe

<q> | blocks_count pipe counts the number of blocks with logs processed by <q>. This pipe is needed mostly for debugging.

See also:

copy pipe

If some log fields must be copied, then | copy src1 as dst1, ..., srcN as dstN pipe can be used. For example, the following query copies host field to server for logs over the last 5 minutes, so the output contains both host and server fields:

_time:5m | copy host as server

Multiple fields can be copied with a single | copy ... pipe. For example, the following query copies _time field to timestamp, while _msg field is copied to message:

_time:5m | copy _time as timestamp, _msg as message

The as keyword is optional.

cp keyword can be used instead of copy for convenience. For example, _time:5m | cp foo bar is equivalent to _time:5m | copy foo as bar.

See also:

delete pipe

If some log fields must be deleted, then | delete field1, ..., fieldN pipe can be used. For example, the following query deletes host and app fields from the logs over the last 5 minutes:

_time:5m | delete host, app

drop, del and rm keywords can be used instead of delete for convenience. For example, _time:5m | drop host is equivalent to _time:5m | delete host.

See also:

drop_empty_fields pipe

| drop_empty_fields pipe drops fields with empty values. It also skips log entries with zero non-empty fields.

For example, the following query drops possible empty email field generated by extract pipe if the foo field doesn't contain email:

_time:5m | extract 'email: <email>,' from foo | drop_empty_fields

See also:

extract pipe

| extract "pattern" from field_name pipe allows extracting arbitrary text into output fields according to the pattern from the given field_name. Existing log fields remain unchanged after the | extract ... pipe.

| extract ... can be useful for extracting additional fields needed for further data processing with other pipes such as stats pipe or sort pipe.

For example, the following query selects logs with the error word for the last day, extracts ip address from _msg field into ip field and then calculates top 10 ip addresses with the biggest number of logs:

_time:1d error | extract "ip=<ip> " from _msg | stats by (ip) count() logs | sort by (logs) desc limit 10

It is expected that _msg field contains ip=... substring ending with space. For example, error ip=1.2.3.4 from user_id=42. If there is no such substring in the current _msg field, then the ip output field will be empty.

If the | extract ... pipe is applied to _msg field, then the from _msg part can be omitted. For example, the following query is equivalent to the previous one:

_time:1d error | extract "ip=<ip> " | stats by (ip) count() logs | sort by (logs) desc limit 10

If the pattern contains double quotes, then either put \ in front of double quotes or put the pattern inside single quotes. For example, the following query extracts ip from the corresponding JSON field:

_time:5m | extract '"ip":"<ip>"'

Add keep_original_fields to the end of extract ... when the original non-empty values of the fields mentioned in the pattern must be preserved instead of overwriting it with the extracted values. For example, the following query extracts <ip> only if the original value for ip field is missing or is empty:

_time:5m | extract 'ip=<ip> ' keep_original_fields

By default extract writes empty matching fields to the output, which may overwrite existing values. Add skip_empty_results to the end of extract ... in order to prevent from overwriting the existing values for the corresponding fields with empty values. For example, the following query preserves the original ip field value if foo field doesn't contain the matching ip:

_time:5m | extract 'ip=<ip> ' from foo skip_empty_results

Performance tip: it is recommended using more specific log filters in order to reduce the number of log entries, which are passed to extract. See general performance tips for details.

See also:

Format for extract pipe pattern

The pattern part from extract pipe has the following format:

text1<field1>text2<field2>...textN<fieldN>textN+1

Where text1, ... textN+1 is arbitrary non-empty text, which matches as is to the input text.

The field1, ... fieldN are placeholders, which match a substring of any length (including zero length) in the input text until the next textX. Placeholders can be anonymous and named. Anonymous placeholders are written as <_>. They are used for convenience when some input text must be skipped until the next textX. Named placeholders are written as <some_name>, where some_name is the name of the log field to store the corresponding matching substring to.

Matching starts from the first occurrence of the text1 in the input text. If the pattern starts with <field1> and doesn't contain text1, then the matching starts from the beginning of the input text. Matching is performed sequentially according to the pattern. If some textX isn't found in the remaining input text, then the remaining named placeholders receive empty string values and the matching finishes prematurely. The empty string values can be dropped with drop_empty_fields pipe.

Matching finishes successfully when textN+1 is found in the input text. If the pattern ends with <fieldN> and doesn't contain textN+1, then the <fieldN> matches the remaining input text.

For example, if _msg field contains the following text:

1.2.3.4 GET /foo/bar?baz 404 "Mozilla  foo bar baz" some tail here

Then the following pattern can be used for extracting ip, path and user_agent fields from it:

<ip> <_> <path> <_> "<user_agent>"

Note that the user-agent part of the log message is in double quotes. This means that it may contain special chars, including escaped double quote, e.g. \". This may break proper matching of the string in double quotes.

VictoriaLogs automatically detects quoted strings and automatically unquotes them if the first matching char in the placeholder is double quote or backtick. So it is better to use the following pattern for proper matching of quoted user_agent string:

<ip> <_> <path> <_> <user_agent>

This is useful for extracting JSON strings. For example, the following pattern properly extracts the message JSON string into msg field, even if it contains special chars:

"message":<msg>

The automatic string unquoting can be disabled if needed by adding plain: prefix in front of the field name. For example, if some JSON array of string values must be captured into json_array field, then the following pattern can be used:

some json string array: [<plain:json_array>]

If some special chars such as < must be matched by the pattern, then they can be html-escaped. For example, the following pattern properly matches a < b text by extracting a into left field and b into right field:

<left> &lt; <right>

Conditional extract

If some log entries must be skipped from extract pipe, then add if (<filters>) filter after the extract word. The <filters> can contain arbitrary filters. For example, the following query extracts ip field from _msg field only if the input log entry doesn't contain ip field or this field is empty:

_time:5m | extract if (ip:"") "ip=<ip> "

An alternative approach is to add keep_original_fields to the end of extract, in order to keep the original non-empty values for the extracted fields. For example, the following query is equivalent to the previous one:

_time:5m | extract "ip=<ip> " keep_original_fields

extract_regexp pipe

| extract_regexp "pattern" from field_name pipe extracts substrings from the field_name field according to the provided pattern, and stores them into field names according to the named fields inside the pattern. The pattern must contain RE2 regular expression with named fields (aka capturing groups) in the form (?P<capture_field_name>...). Matching substrings are stored to the given capture_field_name log fields. For example, the following query extracts ipv4 addresses from _msg field and puts them into ip field for logs over the last 5 minutes:

_time:5m | extract_regexp "(?P<ip>([0-9]+[.]){3}[0-9]+)" from _msg

The from _msg part can be omitted if the data extraction is performed from the _msg field. So the following query is equivalent to the previous one:

_time:5m | extract_regexp "(?P<ip>([0-9]+[.]){3}[0-9]+)"

Add keep_original_fields to the end of extract_regexp ... when the original non-empty values of the fields mentioned in the pattern must be preserved instead of overwriting it with the extracted values. For example, the following query extracts <ip> only if the original value for ip field is missing or is empty:

_time:5m | extract_regexp 'ip=(?P<ip>([0-9]+[.]){3}[0-9]+)' keep_original_fields

By default extract_regexp writes empty matching fields to the output, which may overwrite existing values. Add skip_empty_results to the end of extract_regexp ... in order to prevent from overwriting the existing values for the corresponding fields with empty values. For example, the following query preserves the original ip field value if foo field doesn't contain the matching ip:

_time:5m | extract_regexp 'ip=(?P<ip>([0-9]+[.]){3}[0-9]+)' from foo skip_empty_results

Performance tip: it is recommended using extract pipe instead of extract_regexp for achieving higher query performance.

See also:

Conditional extract_regexp

If some log entries must be skipped from extract_regexp pipe, then add if (<filters>) filter after the extract word. The <filters> can contain arbitrary filters. For example, the following query extracts ip from _msg field only if the input log entry doesn't contain ip field or this field is empty:

_time:5m | extract_regexp if (ip:"") "ip=(?P<ip>([0-9]+[.]){3}[0-9]+)"

An alternative approach is to add keep_original_fields to the end of extract_regexp, in order to keep the original non-empty values for the extracted fields. For example, the following query is equivalent to the previous one:

_time:5m | extract_regexp "ip=(?P<ip>([0-9]+[.]){3}[0-9]+)" keep_original_fields

field_names pipe

| field_names pipe returns all the names of log fields with an estimated number of logs per each field name. For example, the following query returns all the field names with the number of matching logs over the last 5 minutes:

_time:5m | field_names

Field names are returned in arbitrary order. Use sort pipe in order to sort them if needed.

See also:

field_values pipe

| field_values field_name pipe returns all the values for the given field_name field with the number of logs per each value. For example, the following query returns all the values with the number of matching logs for the field level over logs for the last 5 minutes:

_time:5m | field_values level

It is possible limiting the number of returned values by adding limit N to the end of the field_values .... For example, the following query returns up to 10 values for the field user_id over logs for the last 5 minutes:

_time:5m | field_values user_id limit 10

If the limit is reached, then the set of returned values is random. Also the number of matching logs per each returned value is zeroed for performance reasons.

See also:

fields pipe

By default all the log fields are returned in the response. It is possible to select the given set of log fields with | fields field1, ..., fieldN pipe. For example, the following query selects only host and _msg fields from logs for the last 5 minutes:

_time:5m | fields host, _msg

keep can be used instead of fields for convenience. For example, the following query is equivalent to the previous one:

_time:5m | keep host, _msg

See also:

filter pipe

The | filter ... pipe allows filtering the selected logs entries with arbitrary filters.

For example, the following query returns host field values if the number of log messages with the error word for them over the last hour exceeds 1_000:

_time:1h error | stats by (host) count() logs_count | filter logs_count:> 1_000

It is allowed to use where prefix instead of filter prefix for convenience. For example, the following query is equivalent to the previous one:

_time:1h error | stats by (host) count() logs_count | where logs_count:> 1_000

It is allowed to omit filter prefix if the used filters do not clash with pipe names. So the following query is equivalent to the previous one:

_time:1h error | stats by (host) count() logs_count | logs_count:> 1_000

See also:

format pipe

| format "pattern" as result_field pipe combines log fields according to the pattern and stores it to the result_field.

For example, the following query stores request from <ip>:<port> text into _msg field, by substituting <ip> and <port> with the corresponding log field values:

_time:5m | format "request from <ip>:<port>" as _msg

If the result of the format pattern is stored into _msg field, then as _msg part can be omitted. The following query is equivalent to the previous one:

_time:5m | format "request from <ip>:<port>"

If some field values must be put into double quotes before formatting, then add q: in front of the corresponding field name. For example, the following command generates properly encoded JSON object from _msg and stacktrace log fields and stores it into my_json output field:

_time:5m | format '{"_msg":<q:_msg>,"stacktrace":<q:stacktrace>}' as my_json

Numeric fields can be transformed into the following string representation at format pipe:

  • RFC3339 time - by adding time: in front of the corresponding field name containing Unix timestamp in nanoseconds. For example, format "time=<time:timestamp_nsecs>". The timestamp can be converted into nanoseconds with the math pipe.

  • Human-readable duration - by adding duration: in front of the corresponding numeric field name containing duration in nanoseconds. For example, format "duration=<duration:duration_nsecs>". The duration can be converted into nanoseconds with the math pipe.

  • IPv4 - by adding ipv4: in front of the corresponding field name containing uint32 representation of the IPv4 address. For example, format "ip=<ipv4:ip_num>".

Add keep_original_fields to the end of format ... as result_field when the original non-empty value of the result_field must be preserved instead of overwriting it with the format results. For example, the following query adds formatted result to foo field only if it was missing or empty:

_time:5m | format 'some_text' as foo keep_original_fields

Add skip_empty_results to the end of format ... if empty results shouldn't be written to the output. For example, the following query adds formatted result to foo field when at least field1 or field2 aren't empty, while preserving the original foo value:

_time:5m | format "<field1><field2>" as foo skip_empty_results

Performance tip: it is recommended using more specific log filters in order to reduce the number of log entries, which are passed to format. See general performance tips for details.

See also:

Conditional format

If the format pipe mustn't be applied to every log entry, then add if (<filters>) just after the format word. The <filters> can contain arbitrary filters. For example, the following query stores the formatted result to message field only if ip and host fields aren't empty:

_time:5m | format if (ip:* and host:*) "request from <ip>:<host>" as message

join pipe

The | join by (<fields>) (<query>) pipe joins the current results with the <query> results by the given set of comma-separated <fields>. This pipe works in the following way:

  1. It executes the <query> and remembers its' results. It may contain arbitrary LogsQL query.
  2. For each input row it searches for matching rows in the <query> results by the given <fields>.
  3. If the <query> results have no matching rows, then the input row is sent to the output as is.
  4. If the <query> results has matching rows, then for each matching row the input row is extended with new fields seen at the matching row, and the result is sent to the output.

This logic is similar to LEFT JOIN in SQL. For example, the following query returns the number of per-user logs across two applications - app1 and app2 ( see stream filters for details on {...} filter):

_time:1d {app="app1"} | stats by (user) count() app1_hits
  | join by (user) (
    _time:1d {app="app2"} | stats by (user) count() app2_hits
  )

If you need results similar to JOIN in SQL, then apply filter pipe with * filter on fields, which must be non-empty after the join. For example, the following query returns stats only for users, which exist in both applications app1 and app2:

_time:1d {app="app1"} | stats by (user) count() app1_hits
  | join by (user) (
    _time:1d {app="app2"} | stats by (user) count() app2_hits
  )
  | filter app2_hits:*

It is possible adding a prefix to all the field names returned by the <query> by specifying the needed prefix after the <query>. For example, the following query adds app2. prefix to all <query> log fields:

_time:1d {app="app1"} | stats by (user) count() app1_hits
  | join by (user) (
    _time:1d {app="app2"} | stats by (user) count() app2_hits
  ) prefix "app2."

Performance tips:

  • Make sure that the <query> in the join pipe returns relatively small number of results, since they are kept in RAM during execution of join pipe.
  • Conditional stats is usually faster to execute. They usually require less RAM than the equivalent join pipe.

See also:

len pipe

The | len(field) as result pipe stores byte length of the given field value into the result field. For example, the following query shows top 5 log entries with the maximum byte length of _msg field across logs for the last 5 minutes:

_time:5m | len(_msg) as msg_len | sort by (msg_len desc) | limit 1

See also:

limit pipe

If only a subset of selected logs must be processed, then | limit N pipe can be used, where N can contain any supported integer numeric value. For example, the following query returns up to 100 logs over the last 5 minutes:

_time:5m | limit 100

head keyword can be used instead of limit for convenience. For example, _time:5m | head 100 is equivalent to _time:5m | limit 100.

The N in head N can be omitted - in this case up to 10 matching logs are returned:

error | head

By default rows are selected in arbitrary order because of performance reasons, so the query above can return different sets of logs every time it is executed. sort pipe can be used for making sure the logs are in the same order before applying limit ... to them.

See also:

math pipe

| math ... pipe performs mathematical calculations over numeric values stored in log fields. It has the following format:

| math
  expr1 as resultName1,
  ...
  exprN as resultNameN

Where exprX is one of the supported math expressions mentioned below, while resultNameX is the name of the field to store the calculated result to. The as keyword is optional. The result name can be omitted. In this case the result is stored to a field with the name equal to string representation of the corresponding math expression.

exprX may reference resultNameY calculated before the given exprX.

For example, the following query divides duration_msecs field value by 1000, then rounds it to integer and stores the result in the duration_secs field:

_time:5m | math round(duration_msecs / 1000) as duration_secs

The following mathematical operations are supported by math pipe:

  • arg1 + arg2 - returns the sum of arg1 and arg2
  • arg1 - arg2 - returns the difference between arg1 and arg2
  • arg1 * arg2 - multiplies arg1 by arg2
  • arg1 / arg2 - divides arg1 by arg2
  • arg1 % arg2 - returns the remainder of the division of arg1 by arg2
  • arg1 ^ arg2 - returns the power of arg1 by arg2
  • arg1 & arg2 - returns bitwise and for arg1 and arg2. It is expected that arg1 and arg2 are in the range [0 .. 2^53-1]
  • arg1 or arg2 - returns bitwise or for arg1 and arg2. It is expected that arg1 and arg2 are in the range [0 .. 2^53-1]
  • arg1 xor arg2 - returns bitwise xor for arg1 and arg2. It is expected that arg1 and arg2 are in the range [0 .. 2^53-1]
  • arg1 default arg2 - returns arg2 if arg1 is non-numeric or equals to NaN
  • abs(arg) - returns an absolute value for the given arg
  • ceil(arg) - returns the least integer value greater than or equal to arg
  • exp(arg) - powers e by arg
  • floor(arg) - returns the greatest integer values less than or equal to arg
  • ln(arg) - returns natural logarithm for the given arg
  • max(arg1, ..., argN) - returns the maximum value among the given arg1, ..., argN
  • min(arg1, ..., argN) - returns the minimum value among the given arg1, ..., argN
  • round(arg) - returns rounded to integer value for the given arg. The round() accepts optional nearest arg, which allows rounding the number to the given nearest multiple. For example, round(temperature, 0.1) rounds temperature field to one decimal digit after the point.

Every argX argument in every mathematical operation can contain one of the following values:

  • The name of log field. For example, errors_total / requests_total. The log field is parsed into numeric value if it contains supported numeric value. The log field is parsed into Unix timestamp in nanoseconds if it contains rfc3339 time. The log field is parsed into uint32 number if it contains IPv4 address. The log field is parsed into NaN in other cases.
  • Any supported numeric value, rfc3339 time or IPv4 address. For example, 1MiB, "2024-05-15T10:20:30.934324Z" or "12.34.56.78".
  • Another mathematical expression, which can be put inside (...). For example, (a + b) * c.

The parsed time, duration and IPv4 address can be converted back to string representation after math transformations with the help of format pipe. For example, the following query rounds the request_duration field to seconds before converting it back to string representation:

_time:5m | math round(request_duration, 1e9) as request_duration_nsecs | format '<duration:request_duration_nsecs>' as request_duration

The eval keyword can be used instead of math for convenience. For example, the following query calculates duration_msecs field by multiplying duration_secs field to 1000:

_time:5m | eval (duration_secs * 1000) as duration_msecs

See also:

offset pipe

If some selected logs must be skipped after sort, then | offset N pipe can be used, where N can contain any supported integer numeric value. For example, the following query skips the first 100 logs over the last 5 minutes after sorting them by _time:

_time:5m | sort by (_time) | offset 100

skip keyword can be used instead of offset keyword for convenience. For example, _time:5m | skip 10 is equivalent to _time:5m | offset 10.

Note that skipping rows without sorting has little sense, since they can be returned in arbitrary order because of performance reasons. Rows can be sorted with sort pipe.

See also:

pack_json pipe

| pack_json as field_name pipe packs all log fields into JSON object and stores it as a string in the given field_name.

For example, the following query packs all the fields into JSON object and stores it into _msg field for logs over the last 5 minutes:

_time:5m | pack_json as _msg

The as _msg part can be omitted if packed JSON object is stored into _msg field. The following query is equivalent to the previous one:

_time:5m | pack_json

If only a subset of labels must be packed into JSON, then it must be listed inside fields (...) after pack_json. For example, the following query builds JSON with foo and bar fields only and stores the result in baz field:

_time:5m | pack_json fields (foo, bar) as baz

The pack_json doesn't modify or delete other labels. If you do not need them, then add | fields ... after the pack_json pipe. For example, the following query leaves only the foo label with the original log fields packed into JSON:

_time:5m | pack_json as foo | fields foo

See also:

pack_logfmt pipe

| pack_logfmt as field_name pipe packs all log fields into logfmt message and stores it as a string in the given field_name.

For example, the following query packs all the fields into logfmt message and stores it into _msg field for logs over the last 5 minutes:

_time:5m | pack_logfmt as _msg

The as _msg part can be omitted if packed message is stored into _msg field. The following query is equivalent to the previous one:

_time:5m | pack_logfmt

If only a subset of labels must be packed into logfmt, then it must be listed inside fields (...) after pack_logfmt. For example, the following query builds logfmt message with foo and bar fields only and stores the result in baz field:

_time:5m | pack_logfmt fields (foo, bar) as baz

The pack_logfmt doesn't modify or delete other labels. If you do not need them, then add | fields ... after the pack_logfmt pipe. For example, the following query leaves only the foo label with the original log fields packed into logfmt:

_time:5m | pack_logfmt as foo | fields foo

See also:

rename pipe

If some log fields must be renamed, then | rename src1 as dst1, ..., srcN as dstN pipe can be used. For example, the following query renames host field to server for logs over the last 5 minutes, so the output contains server field instead of host field:

_time:5m | rename host as server

Multiple fields can be renamed with a single | rename ... pipe. For example, the following query renames host to instance and app to job:

_time:5m | rename host as instance, app as job

The as keyword is optional.

mv keyword can be used instead of rename keyword for convenience. For example, _time:5m | mv foo bar is equivalent to _time:5m | rename foo as bar.

See also:

replace pipe

| replace ("old", "new") at field pipe replaces all the occurrences of the old substring with the new substring in the given field.

For example, the following query replaces all the secret-password substrings with *** in the _msg field for logs over the last 5 minutes:

_time:5m | replace ("secret-password", "***") at _msg

The at _msg part can be omitted if the replacement occurs in the _msg field. The following query is equivalent to the previous one:

_time:5m | replace ("secret-password", "***")

The number of replacements can be limited with limit N at the end of replace. For example, the following query replaces only the first foo substring with bar at the log field baz:

_time:5m | replace ('foo', 'bar') at baz limit 1

Performance tip: it is recommended using more specific log filters in order to reduce the number of log entries, which are passed to replace. See general performance tips for details.

See also:

Conditional replace

If the replace pipe mustn't be applied to every log entry, then add if (<filters>) after replace. The <filters> can contain arbitrary filters. For example, the following query replaces secret with *** in the password field only if user_type field equals to admin:

_time:5m | replace if (user_type:=admin) ("secret", "***") at password

replace_regexp pipe

| replace_regexp ("regexp", "replacement") at field pipe replaces all the substrings matching the given regexp with the given replacement in the given field.

The regexp must contain regular expression with RE2 syntax. The replacement may contain $N or ${N} placeholders, which are substituted with the N-th capturing group in the regexp.

For example, the following query replaces all the substrings starting with host- and ending with -foo with the contents between host- and -foo in the _msg field for logs over the last 5 minutes:

_time:5m | replace_regexp ("host-(.+?)-foo", "$1") at _msg

The at _msg part can be omitted if the replacement occurs in the _msg field. The following query is equivalent to the previous one:

_time:5m | replace_regexp ("host-(.+?)-foo", "$1")

The number of replacements can be limited with limit N at the end of replace. For example, the following query replaces only the first password: ... substring ending with whitespace with empty substring at the log field baz:

_time:5m | replace_regexp ('password: [^ ]+', '') at baz limit 1

Performance tips:

  • It is recommended using replace pipe instead of replace_regexp if possible, since it works faster.
  • It is recommended using more specific log filters in order to reduce the number of log entries, which are passed to replace. See general performance tips for details.

See also:

Conditional replace_regexp

If the replace_regexp pipe mustn't be applied to every log entry, then add if (<filters>) after replace_regexp. The <filters> can contain arbitrary filters. For example, the following query replaces password: ... substrings ending with whitespace with *** in the foo field only if user_type field equals to admin:

_time:5m | replace_regexp if (user_type:=admin) ("password: [^ ]+", "") at foo

sort pipe

By default logs are selected in arbitrary order because of performance reasons. If logs must be sorted, then | sort by (field1, ..., fieldN) pipe can be used. The returned logs are sorted by the given fields using natural sorting.

For example, the following query returns logs for the last 5 minutes sorted by _stream and then by _time:

_time:5m | sort by (_stream, _time)

Add desc after the given log field in order to sort in reverse order of this field. For example, the following query sorts log fields in reverse order of request_duration_seconds field:

_time:5m | sort by (request_duration_seconds desc)

The reverse order can be applied globally via desc keyword after by(...) clause:

_time:5m | sort by (foo, bar) desc

The by keyword can be skipped in sort ... pipe. For example, the following query is equivalent to the previous one:

_time:5m | sort (foo, bar) desc

The order alias can be used instead of sort, so the following query is equivalent to the previous one:

_time:5m | order by (foo, bar) desc

Sorting of big number of logs can consume a lot of CPU time and memory. Sometimes it is enough to return the first N entries with the biggest or the smallest values. This can be done by adding limit N to the end of sort ... pipe. Such a query consumes lower amounts of memory when sorting big number of logs, since it keeps in memory only N log entries. For example, the following query returns top 10 log entries with the biggest values for the request_duration field during the last hour:

_time:1h | sort by (request_duration desc) limit 10

If the first N sorted results must be skipped, then offset N can be added to sort pipe. For example, the following query skips the first 10 logs with the biggest request_duration field, and then returns the next 20 sorted logs for the last 5 minutes:

_time:1h | sort by (request_duration desc) offset 10 limit 20

It is possible returning a rank (sort order number) for every sorted log by adding rank as <fieldName> to the end of | sort ... pipe. For example, the following query stores rank for sorted by _time logs into position field:

_time:5m | sort by (_time) rank as position

Note that sorting of big number of logs can be slow and can consume a lot of additional memory. It is recommended limiting the number of logs before sorting with the following approaches:

  • Adding limit N to the end of sort ... pipe.
  • Reducing the selected time range with time filter.
  • Using more specific filters, so they select less logs.
  • Limiting the number of selected fields via fields pipe.

See also:

stats pipe

| stats ... pipe allows calculating various stats over the selected logs. For example, the following LogsQL query uses count stats function for calculating the number of logs for the last 5 minutes:

_time:5m | stats count() as logs_total

| stats ... pipe has the following basic format:

... | stats
  stats_func1(...) as result_name1,
  ...
  stats_funcN(...) as result_nameN

Where stats_func* is any of the supported stats function, while result_name* is the name of the log field to store the result of the corresponding stats function. The as keyword is optional.

For example, the following query calculates the following stats for logs over the last 5 minutes:

_time:5m | stats count() logs_total, count_uniq(_stream) streams_total

It is allowed omitting stats prefix for convenience. So the following query is equivalent to the previous one:

_time:5m | count() logs_total, count_uniq(_stream) streams_total

It is allowed omitting the result name. In this case the result name equals to the string representation of the used stats function. For example, the following query returns the same stats as the previous one, but gives uses count() and count_uniq(_stream) names for the returned fields:

_time:5m | count(), count_uniq(_stream)

See also:

Stats by fields

The following LogsQL syntax can be used for calculating independent stats per group of log fields:

... | stats by (field1, ..., fieldM)
  stats_func1(...) as result_name1,
  ...
  stats_funcN(...) as result_nameN

This calculates stats_func* per each (field1, ..., fieldM) group of log fields.

For example, the following query calculates the number of logs and unique ip addresses over the last 5 minutes, grouped by (host, path) fields:

_time:5m | stats by (host, path) count() logs_total, count_uniq(ip) ips_total

The by keyword can be skipped in stats ... pipe. For example, the following query is equivalent to the previous one:

_time:5m | stats (host, path) count() logs_total, count_uniq(ip) ips_total

See also:

Stats by time buckets

The following syntax can be used for calculating stats grouped by time buckets:

... | stats by (_time:step)
  stats_func1(...) as result_name1,
  ...
  stats_funcN(...) as result_nameN

This calculates stats_func* per each step of _time field. The step can have any duration value. For example, the following LogsQL query returns per-minute number of logs and unique ip addresses over the last 5 minutes:

_time:5m | stats by (_time:1m) count() logs_total, count_uniq(ip) ips_total

Additionally, the following step values are supported:

  • nanosecond - equals to 1ns duration.
  • microsecond - equals to 1µs duration.
  • millisecond - equals to 1ms duration.
  • second - equals to 1s duration.
  • minute - equals to 1m duration.
  • hour - equals to 1h duration.
  • day - equals to 1d duration.
  • week - equals to 1w duration.
  • month - equals to one month. It properly takes into account the number of days per each month.
  • year - equals to one year. It properly takes into account the number of days per each year.

Stats by time buckets with timezone offset

VictoriaLogs stores _time values as Unix time in nanoseconds. This time corresponds to UTC time zone. Sometimes it is needed calculating stats grouped by days or weeks at non-UTC timezone. This is possible with the following syntax:

... | stats by (_time:step offset timezone_offset) ...

For example, the following query calculates per-day number of logs over the last week, in UTC+02:00 time zone:

_time:1w | stats by (_time:1d offset 2h) count() logs_total

Stats by field buckets

Every log field inside | stats by (...) can be bucketed in the same way at _time field in this example. Any numeric value can be used as step value for the bucket. For example, the following query calculates the number of requests for the last hour, bucketed by 10KB of request_size_bytes field:

_time:1h | stats by (request_size_bytes:10KB) count() requests

Stats by IPv4 buckets

Stats can be bucketed by log field containing IPv4 addresses via the ip_field_name:/network_mask syntax inside by(...) clause. For example, the following query returns the number of log entries per /24 subnetwork extracted from the ip log field during the last 5 minutes:

_time:5m | stats by (ip:/24) count() requests_per_subnet

Stats with additional filters

Sometimes it is needed to calculate stats on different subsets of matching logs. This can be done by inserting if (<any_filters>) condition between stats function and result_name, where any_filter can contain arbitrary filters. For example, the following query calculates individually the number of logs messages with GET, POST and PUT words, additionally to the total number of logs over the last 5 minutes:

_time:5m | stats
  count() if (GET) gets,
  count() if (POST) posts,
  count() if (PUT) puts,
  count() total

If zero input rows match the given if (...) filter, then zero result is returned for the given stats function.

See also:

stream_context pipe

| stream_context ... pipe allows selecting surrounding logs for the matching logs in logs stream in the way similar to grep -A / grep -B. The returned log chunks are delimited with --- log message for easier investigation.

For example, the following query returns up to 10 additional logs after every log message with the panic word across all the logs for the last 5 minutes:

_time:5m panic | stream_context after 10

The following query returns up to 5 additional logs in front of every log message with the stacktrace word across all the logs for the last 5 minutes:

_time:5m stacktrace | stream_context before 5

The following query returns up to 2 logs in front of the log message with the error word and up to 5 logs after this log message across all the logs for the last 5 minutes:

_time:5m error | stream_context before 2 after 5

The | stream_context pipe must go first just after the filters.

top pipe

| top N by (field1, ..., fieldN) pipe returns top N sets for (field1, ..., fieldN) log fields with the maximum number of matching log entries.

For example, the following query returns top 7 log streams with the maximum number of log entries over the last 5 minutes:

_time:5m | top 7 by (_stream)

The N is optional. If it is skipped, then top 10 entries are returned. For example, the following query returns top 10 values for ip field seen in logs for the last 5 minutes:

_time:5m | top by (ip)

The by (...) part in the top pipe is optional. If it is skipped, then all the log fields are taken into account when determining top field sets. This is useful when the field sets are already limited by other pipes such as fields pipe. For example, the following query is equivalent to the previous one:

_time:5m | fields ip | top

It is possible to set rank field per each returned entry for top pipe by adding with rank. For example, the following query sets the rank field per each returned ip:

_time:5m | top 10 by (ip) rank

The rank field can have other name. For example, the following query uses the position field name instead of rank field name in the output:

_time:5m | top 10 by (ip) rank as position

See also:

uniq pipe

| uniq ... pipe returns unique results over the selected logs. For example, the following LogsQL query returns unique values for ip log field over logs for the last 5 minutes:

_time:5m | uniq by (ip)

It is possible to specify multiple fields inside by(...) clause. In this case all the unique sets for the given fields are returned. For example, the following query returns all the unique (host, path) pairs for the logs over the last 5 minutes:

_time:5m | uniq by (host, path)

The unique entries are returned in arbitrary order. Use sort pipe in order to sort them if needed.

Add with hits after uniq by (...) in order to return the number of matching logs per each field value:

_time:5m | uniq by (host) with hits

Unique entries are stored in memory during query execution. Big number of unique selected entries may require a lot of memory. Sometimes it is enough to return up to N unique entries. This can be done by adding limit N after by (...) clause. This allows limiting memory usage. For example, the following query returns up to 100 unique (host, path) pairs for the logs over the last 5 minutes:

_time:5m | uniq by (host, path) limit 100

If the limit is reached, then arbitrary subset of unique values can be returned. The hits calculation doesn't work when the limit is reached.

The by keyword can be skipped in uniq ... pipe. For example, the following query is equivalent to the previous one:

_time:5m | uniq (host, path) limit 100

See also:

unpack_json pipe

| unpack_json from field_name pipe unpacks {"k1":"v1", ..., "kN":"vN"} JSON from the given input field_name into k1, ... kN output field names with the corresponding v1, ..., vN values. It overrides existing fields with names from the k1, ..., kN list. Other fields remain untouched.

Nested JSON is unpacked according to the rules defined here.

For example, the following query unpacks JSON fields from the _msg field across logs for the last 5 minutes:

_time:5m | unpack_json from _msg

The from _msg part can be omitted when JSON fields are unpacked from the _msg field. The following query is equivalent to the previous one:

_time:5m | unpack_json

If only some fields must be extracted from JSON, then they can be enumerated inside fields (...). For example, the following query unpacks only foo and bar fields from JSON value stored in my_json log field:

_time:5m | unpack_json from my_json fields (foo, bar)

If it is needed to preserve the original non-empty field values, then add keep_original_fields to the end of unpack_json .... For example, the following query preserves the original non-empty values for ip and host fields instead of overwriting them with the unpacked values:

_time:5m | unpack_json from foo fields (ip, host) keep_original_fields

Add skip_empty_results to the end of unpack_json ... if the original field values must be preserved when the corresponding unpacked values are empty. For example, the following query preserves the original ip and host field values for empty unpacked values:

_time:5m | unpack_json fields (ip, host) skip_empty_results

Performance tip: if you need extracting a single field from long JSON, it is faster to use extract pipe. For example, the following query extracts "ip" field from JSON stored in _msg field at the maximum speed:

_time:5m | extract '"ip":<ip>'

If you want to make sure that the unpacked JSON fields do not clash with the existing fields, then specify common prefix for all the fields extracted from JSON, by adding result_prefix "prefix_name" to unpack_json. For example, the following query adds foo_ prefix for all the unpacked fields form foo:

_time:5m | unpack_json from foo result_prefix "foo_"

Performance tips:

  • It is better from performance and resource usage PoV ingesting parsed JSON logs into VictoriaLogs according to the supported data model instead of ingesting unparsed JSON lines into VictoriaLogs and then parsing them at query time with unpack_json pipe.

  • It is recommended using more specific log filters in order to reduce the number of log entries, which are passed to unpack_json. See general performance tips for details.

See also:

Conditional unpack_json

If the unpack_json pipe mustn't be applied to every log entry, then add if (<filters>) after unpack_json. The <filters> can contain arbitrary filters. For example, the following query unpacks JSON fields from foo field only if ip field in the current log entry isn't set or empty:

_time:5m | unpack_json if (ip:"") from foo

unpack_logfmt pipe

| unpack_logfmt from field_name pipe unpacks k1=v1 ... kN=vN logfmt fields from the given field_name into k1, ... kN field names with the corresponding v1, ..., vN values. It overrides existing fields with names from the k1, ..., kN list. Other fields remain untouched.

For example, the following query unpacks logfmt fields from the _msg field across logs for the last 5 minutes:

_time:5m | unpack_logfmt from _msg

The from _msg part can be omitted when logfmt fields are unpacked from the _msg field. The following query is equivalent to the previous one:

_time:5m | unpack_logfmt

If only some fields must be unpacked from logfmt, then they can be enumerated inside fields (...). For example, the following query extracts only foo and bar fields from logfmt stored in the my_logfmt field:

_time:5m | unpack_logfmt from my_logfmt fields (foo, bar)

If it is needed to preserve the original non-empty field values, then add keep_original_fields to the end of unpack_logfmt .... For example, the following query preserves the original non-empty values for ip and host fields instead of overwriting them with the unpacked values:

_time:5m | unpack_logfmt from foo fields (ip, host) keep_original_fields

Add skip_empty_results to the end of unpack_logfmt ... if the original field values must be preserved when the corresponding unpacked values are empty. For example, the following query preserves the original ip and host field values for empty unpacked values:

_time:5m | unpack_logfmt fields (ip, host) skip_empty_results

Performance tip: if you need extracting a single field from long logfmt line, it is faster to use extract pipe. For example, the following query extracts "ip" field from logfmt line stored in _msg field:

_time:5m | extract ' ip=<ip>'

If you want to make sure that the unpacked logfmt fields do not clash with the existing fields, then specify common prefix for all the fields extracted from logfmt, by adding result_prefix "prefix_name" to unpack_logfmt. For example, the following query adds foo_ prefix for all the unpacked fields from foo field:

_time:5m | unpack_logfmt from foo result_prefix "foo_"

Performance tips:

  • It is better from performance and resource usage PoV ingesting parsed logfmt logs into VictoriaLogs according to the supported data model instead of ingesting unparsed logfmt lines into VictoriaLogs and then parsing them at query time with unpack_logfmt pipe.

  • It is recommended using more specific log filters in order to reduce the number of log entries, which are passed to unpack_logfmt. See general performance tips for details.

See also:

Conditional unpack_logfmt

If the unpack_logfmt pipe mustn't be applied to every log entry, then add if (<filters>) after unpack_logfmt. The <filters> can contain arbitrary filters. For example, the following query unpacks logfmt fields from foo field only if ip field in the current log entry isn't set or empty:

_time:5m | unpack_logfmt if (ip:"") from foo

unpack_syslog pipe

| unpack_syslog from field_name pipe unpacks syslog message from the given field_name. It understands the following Syslog formats:

  • RFC3164 aka <PRI>MMM DD hh:mm:ss HOSTNAME APP-NAME[PROCID]: MESSAGE
  • RFC5424 aka <PRI>1 TIMESTAMP HOSTNAME APP-NAME PROCID MSGID [STRUCTURED-DATA] MESSAGE

The following fields are unpacked:

  • priority - it is obtained from PRI.
  • facility - it is calculated as PRI / 8.
  • severity - it is calculated as PRI % 8.
  • format - either rfc3164 or rfc5424 depending on which Syslog format is unpacked.
  • timestamp - timestamp in ISO8601 format. The MMM DD hh:mm:ss timestamp in RFC3164 is automatically converted into ISO8601 format by assuming that the timestamp belongs to the last 12 months.
  • hostname
  • app_name
  • proc_id
  • msg_id
  • message

The <PRI> part is optional. If it is missing, then priority, facility and severity fields aren't set.

The [STRUCTURED-DATA] is parsed into fields with the SD-ID.param1, SD-ID.param2, ..., SD-ID.paramN names and the corresponding values according to the specification.

For example, the following query unpacks syslog message from the _msg field across logs for the last 5 minutes:

_time:5m | unpack_syslog from _msg

The from _msg part can be omitted when syslog message is unpacked from the _msg field. The following query is equivalent to the previous one:

_time:5m | unpack_syslog

By default timestamps in RFC3164 format are converted to local timezone. It is possible to change the timezone offset via offset option. For example, the following query adds 5 hours and 30 minutes to unpacked rfc3164 timestamps:

_time:5m | unpack_syslog offset 5h30m

If it is needed to preserve the original non-empty field values, then add keep_original_fields to the end of unpack_syslog ...:

_time:5m | unpack_syslog keep_original_fields

If you want to make sure that the unpacked syslog fields do not clash with the existing fields, then specify common prefix for all the fields extracted from syslog, by adding result_prefix "prefix_name" to unpack_syslog. For example, the following query adds foo_ prefix for all the unpacked fields from foo field:

_time:5m | unpack_syslog from foo result_prefix "foo_"

Performance tips:

  • It is better from performance and resource usage PoV ingesting parsed syslog messages into VictoriaLogs according to the supported data model instead of ingesting unparsed syslog lines into VictoriaLogs and then parsing them at query time with unpack_syslog pipe.

  • It is recommended using more specific log filters in order to reduce the number of log entries, which are passed to unpack_syslog. See general performance tips for details.

See also:

Conditional unpack_syslog

If the unpack_syslog pipe mustn't be applied to every log entry, then add if (<filters>) after unpack_syslog. The <filters> can contain arbitrary filters. For example, the following query unpacks syslog message fields from foo field only if hostname field in the current log entry isn't set or empty:

_time:5m | unpack_syslog if (hostname:"") from foo

unroll pipe

| unroll by (field1, ..., fieldN) pipe can be used for unrolling JSON arrays from field1, fieldN log fields into separate rows.

For example, the following query unrolls timestamp and value log fields from logs for the last 5 minutes:

_time:5m | unroll (timestamp, value)

See also:

Conditional unroll

If the unroll pipe mustn't be applied to every log entry, then add if (<filters>) after unroll. The <filters> can contain arbitrary filters. For example, the following query unrolls value field only if value_type field equals to json_array:

_time:5m | unroll if (value_type:="json_array") (value)

stats pipe functions

LogsQL supports the following functions for stats pipe:

avg stats

avg(field1, ..., fieldN) stats pipe function calculates the average value across all the mentioned log fields. Non-numeric values are ignored.

For example, the following query returns the average value for the duration field over logs for the last 5 minutes:

_time:5m | stats avg(duration) avg_duration

See also:

count stats

count() stats pipe function calculates the number of selected logs.

For example, the following query returns the number of logs over the last 5 minutes:

_time:5m | stats count() logs

It is possible calculating the number of logs with non-empty values for some log field with the count(fieldName) syntax. For example, the following query returns the number of logs with non-empty username field over the last 5 minutes:

_time:5m | stats count(username) logs_with_username

If multiple fields are enumerated inside count(), then it counts the number of logs with at least a single non-empty field mentioned inside count(). For example, the following query returns the number of logs with non-empty username or password fields over the last 5 minutes:

_time:5m | stats count(username, password) logs_with_username_or_password

See also:

count_empty stats

count_empty(field1, ..., fieldN) stats pipe function calculates the number of logs with empty (field1, ..., fieldN) tuples.

For example, the following query calculates the number of logs with empty username field during the last 5 minutes:

_time:5m | stats count_empty(username) logs_with_missing_username

See also:

count_uniq stats

count_uniq(field1, ..., fieldN) stats pipe function calculates the number of unique non-empty (field1, ..., fieldN) tuples.

For example, the following query returns the number of unique non-empty values for ip field over the last 5 minutes:

_time:5m | stats count_uniq(ip) ips

The following query returns the number of unique (host, path) pairs for the corresponding fields over the last 5 minutes:

_time:5m | stats count_uniq(host, path) unique_host_path_pairs

Every unique value is stored in memory during query execution. Big number of unique values may require a lot of memory. Sometimes it is needed to know whether the number of unique values reaches some limit. In this case add limit N just after count_uniq(...) for limiting the number of counted unique values up to N, while limiting the maximum memory usage. For example, the following query counts up to 1_000_000 unique values for the ip field:

_time:5m | stats count_uniq(ip) limit 1_000_000 as ips_1_000_000

See also:

max stats

max(field1, ..., fieldN) stats pipe function returns the maximum value across all the mentioned log fields.

For example, the following query returns the maximum value for the duration field over logs for the last 5 minutes:

_time:5m | stats max(duration) max_duration

row_max function can be used for obtaining other fields with the maximum duration.

See also:

median stats

median(field1, ..., fieldN) stats pipe function calculates the median value across the give numeric log fields.

For example, the following query return median for the duration field over logs for the last 5 minutes:

_time:5m | stats median(duration) median_duration

See also:

min stats

min(field1, ..., fieldN) stats pipe function returns the minimum value across all the mentioned log fields.

For example, the following query returns the minimum value for the duration field over logs for the last 5 minutes:

_time:5m | stats min(duration) min_duration

row_min function can be used for obtaining other fields with the minimum duration.

See also:

quantile stats

quantile(phi, field1, ..., fieldN) stats pipe function calculates phi percentile over numeric values for the given log fields. The phi must be in the range 0 ... 1, where 0 means 0th percentile, while 1 means 100th percentile.

For example, the following query calculates 50th, 90th and 99th percentiles for the request_duration_seconds field over logs for the last 5 minutes:

_time:5m | stats
  quantile(0.5, request_duration_seconds) p50,
  quantile(0.9, request_duration_seconds) p90,
  quantile(0.99, request_duration_seconds) p99

See also:

row_any stats

row_any() stats pipe function returns arbitrary log entry (aka sample) per each selected stats group. Log entry is returned as JSON-encoded dictionary with all the fields from the original log.

For example, the following query returns a sample log entry per each _stream across logs for the last 5 minutes:

_time:5m | stats by (_stream) row_any() as sample_row

Fields from the returned values can be decoded with unpack_json or extract pipes.

If only the specific fields are needed, then they can be enumerated inside row_any(...). For example, the following query returns only _time and path fields from a sample log entry for logs over the last 5 minutes:

_time:5m | stats row_any(_time, path) as time_and_path_sample

See also:

row_max stats

row_max(field) stats pipe function returns log entry with the maximum value for the given field. Log entry is returned as JSON-encoded dictionary with all the fields from the original log.

For example, the following query returns log entry with the maximum value for the duration field across logs for the last 5 minutes:

_time:5m | stats row_max(duration) as log_with_max_duration

Fields from the returned values can be decoded with unpack_json or extract pipes.

If only the specific fields are needed from the returned log entry, then they can be enumerated inside row_max(...). For example, the following query returns only _time, path and duration fields from the log entry with the maximum duration over the last 5 minutes:

_time:5m | stats row_max(duration, _time, path, duration) as time_and_path_with_max_duration

See also:

row_min stats

row_min(field) stats pipe function returns log entry with the minimum value for the given field. Log entry is returned as JSON-encoded dictionary with all the fields from the original log.

For example, the following query returns log entry with the minimum value for the duration field across logs for the last 5 minutes:

_time:5m | stats row_min(duration) as log_with_min_duration

Fields from the returned values can be decoded with unpack_json or extract pipes.

If only the specific fields are needed from the returned log entry, then they can be enumerated inside row_max(...). For example, the following query returns only _time, path and duration fields from the log entry with the minimum duration over the last 5 minutes:

_time:5m | stats row_min(duration, _time, path, duration) as time_and_path_with_min_duration

See also:

sum stats

sum(field1, ..., fieldN) stats pipe function calculates the sum of numeric values across all the mentioned log fields.

For example, the following query returns the sum of numeric values for the duration field over logs for the last 5 minutes:

_time:5m | stats sum(duration) sum_duration

See also:

sum_len stats

sum_len(field1, ..., fieldN) stats pipe function calculates the sum of byte lengths of all the values for the given log fields.

For example, the following query returns the sum of byte lengths of _msg fields across all the logs for the last 5 minutes:

_time:5m | stats sum_len(_msg) messages_len

See also:

uniq_values stats

uniq_values(field1, ..., fieldN) stats pipe function returns the unique non-empty values across the mentioned log fields. The returned values are encoded in sorted JSON array.

For example, the following query returns unique non-empty values for the ip field over logs for the last 5 minutes:

_time:5m | stats uniq_values(ip) unique_ips

The returned unique ip addresses can be unrolled into distinct log entries with unroll pipe.

Every unique value is stored in memory during query execution. Big number of unique values may require a lot of memory. Sometimes it is enough to return only a subset of unique values. In this case add limit N after uniq_values(...) in order to limit the number of returned unique values to N, while limiting the maximum memory usage. For example, the following query returns up to 100 unique values for the ip field over the logs for the last 5 minutes:

_time:5m | stats uniq_values(ip) limit 100 as unique_ips_100

Arbitrary subset of unique ip values is returned every time if the limit is reached.

See also:

values stats

values(field1, ..., fieldN) stats pipe function returns all the values (including empty values) for the mentioned log fields. The returned values are encoded in JSON array.

For example, the following query returns all the values for the ip field over logs for the last 5 minutes:

_time:5m | stats values(ip) ips

The returned ip addresses can be unrolled into distinct log entries with unroll pipe.

See also:

Stream context

See stream_context pipe.

Transformations

LogsQL supports the following transformations on the log entries selected with filters:

It is also possible to perform various transformations on the selected log entries at client side with jq, awk, cut, etc. Unix commands according to these docs.

Post-filters

Post-filtering of query results can be performed at any step by using filter pipe.

It is also possible to perform post-filtering of the selected log entries at client side with grep and similar Unix commands according to these docs.

Stats

Stats over the selected logs can be calculated via stats pipe.

It is also possible to perform stats calculations on the selected log entries at client side with sort, uniq, etc. Unix commands according to these docs.

Sorting

By default VictoriaLogs doesn't sort the returned results because of performance reasons. Use sort pipe for sorting the results.

Limiters

LogsQL provides the following pipes for limiting the number of returned log entries:

Querying specific fields

Specific log fields can be queried via fields pipe.

Comments

LogsQL query may contain comments at any place. The comment starts with # and continues until the end of the current line. Example query with comments:

error                       # find logs with `error` word
  | stats by (_stream) logs # then count the number of logs per `_stream` label
  | sort by (logs) desc     # then sort by the found logs in descending order
  | limit 5                 # and show top 5 streams with the biggest number of logs

Numeric values

LogsQL accepts numeric values in the following formats:

Short numeric values

LogsQL accepts integer and floating point values with the following suffixes:

  • K and KB - the value is multiplied by 10^3
  • M and MB - the value is multiplied by 10^6
  • G and GB - the value is multiplied by 10^9
  • T and TB - the value is multiplied by 10^12
  • Ki and KiB - the value is multiplied by 2^10
  • Mi and MiB - the value is multiplied by 2^20
  • Gi and GiB - the value is multiplied by 2^30
  • Ti and TiB - the value is multiplied by 2^40

All the numbers may contain _ delimiters, which may improve readability of the query. For example, 1_234_567 is equivalent to 1234567, while 1.234_567 is equivalent to 1.234567.

Duration values

LogsQL accepts duration values with the following suffixes at places where the duration is allowed:

  • ns - nanoseconds. For example, 123ns.
  • µs - microseconds. For example, 1.23µs.
  • ms - milliseconds. For example, 1.23456ms
  • s - seconds. For example, 1.234s
  • m - minutes. For example, 1.5m
  • h - hours. For example, 1.5h
  • d - days. For example, 1.5d
  • w - weeks. For example, 1w
  • y - years as 365 days. For example, 1.5y

Multiple durations can be combined. For example, 1h33m55s.

Internally duration values are converted into nanoseconds.

Performance tips

  • It is highly recommended specifying time filter in order to narrow down the search to specific time range.
  • It is highly recommended specifying stream filter in order to narrow down the search to specific log streams.
  • Move faster filters such as word filter and phrase filter to the beginning of the query. This rule doesn't apply to time filter and stream filter, which can be put at any place of the query.
  • Move more specific filters, which match lower number of log entries, to the beginning of the query. This rule doesn't apply to time filter and stream filter, which can be put at any place of the query.
  • If the selected logs are passed to pipes for further transformations and statistics' calculations, then it is recommended reducing the number of selected logs by using more specific filters, which return lower number of logs to process by pipes.