From 894416b4ca757fadb62061894c693ee2163eb0a5 Mon Sep 17 00:00:00 2001 From: hagen1778 Date: Wed, 9 Mar 2022 17:46:40 +0200 Subject: [PATCH] docs: add update details for some releases Some of the releases could negatively affect performance for a limited period of time due to some changes in core. Update details are meant to warn users about expected changes in peformance after the update. Signed-off-by: hagen1778 --- docs/CHANGELOG.md | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+) diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md index 1a594224b..0d9397dff 100644 --- a/docs/CHANGELOG.md +++ b/docs/CHANGELOG.md @@ -69,6 +69,17 @@ Released at 22-02-2022 Released at 14-02-2022 +**Update notes:** changes introduced in [#1401](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1401) +are heavily related to the TSID cache, which is mostly utilized during the ingestion. The fix implies the cache +internal structure change which would cause the cache reset on startup. After the update, users need to expect higher +CPU and memory usage during the ingestion while the cache is being populated. Normally, we expect this process to take +less than hour. + +We recommend updating in "off-peak" time when load on the VictoriaMetrics is on its minimum. We also recommend +skipping [v1.73.0](https://github.com/VictoriaMetrics/VictoriaMetrics/releases/tag/v1.73.0) in favour of +[v1.73.1](https://github.com/VictoriaMetrics/VictoriaMetrics/releases/tag/v1.73.1) which contains the fix for reducing +the memory usage after the update. + * FEATURE: publish VictoriaMetrics binaries for MacOS amd64 and MacOS arm64 (aka MacBook M1) at [releases page](https://github.com/VictoriaMetrics/VictoriaMetrics/releases). See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1896) and [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1851). * FEATURE: reduce CPU and disk IO usage during `indexdb` rotation once per `-retentionPeriod`. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1401). * FEATURE: [VictoriaMetrics cluster](https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html): add `-dropSamplesOnOverload` command-line flag for `vminsert`. If this flag is set, then `vminsert` drops incoming data if the destination `vmstorage` is temporarily unavailable or cannot keep up with the ingestion rate. The number of dropped rows can be [monitored](https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#monitoring) via `vm_rpc_rows_dropped_on_overload_total` metric at `vminsert`. @@ -151,6 +162,13 @@ Released at 18-01-2022 Released at 20-12-2021 +**Update notes:** deduplication logic was slightly changed on the release, which may cause extra +[background merges](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282) +for already existing data parts. This process is intentionally limited by one CPU core, but still can result +into increase of CPU usage until merges are finished. + +We recommend updating in "off-peak" time when load on the VictoriaMetrics is on its minimum. + * FEATURE: [VictoriaMetrics enterprise](https://victoriametrics.com/products/enterprise/): add multi-level downsampling support. See [these docs](https://docs.victoriametrics.com/#downsampling) and [this feature request](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/36). * FEATURE: [vmui](https://docs.victoriametrics.com/#vmui): add ability to analyze the correlation between two queries on a single graph. Just click `+Query` button, enter the second query in the newly appeared input field and press `Ctrl+Enter`. Results for both queries should be displayed simultaneously on the same graph. Every query has its own vertical scale, which is displayed on the left and the right side of the graph. Lines for the second query are dashed. See [this pull request](https://github.com/VictoriaMetrics/VictoriaMetrics/pull/1916). * FEATURE: [vmui](https://docs.victoriametrics.com/#vmui): add ability to override the interval between returned datapoints. By default it is automatically calculated depending on the selected time range and horizontal resolution of the graph. Now it is possible to override it with custom values. This may be useful during data exploration and debugging.