tsdb: Skip clean series during periodic head chunk mmap by aknuds1 · Pull Request #18272 · prometheus/prometheus

aknuds1 · 2026-03-11T07:40:17Z

The periodic mmapHeadChunks cycle previously scanned every series across all stripes, acquiring a per-series lock on each, even though typically >99% have nothing to mmap. Profiling identified this as an ingester CPU bottleneck in Grafana Mimir.

This PR adds a headChunkCount field (sync/atomic.Uint32, 4 bytes) to memSeries that tracks the number of head chunks. It is incremented in cutNewHeadChunk and the histogram new-chunk paths, and reset by mmapChunks and truncateChunksBefore. The mmapHeadChunks function uses a lock-free Load() to skip series with fewer than 2 head chunks — only series with chunks ready for mmapping acquire the series mutex.

sync/atomic.Uint32 (4 bytes, align 4) is used instead of go.uber.org/atomic (8 bytes due to nocmp) to fit in existing struct padding without growing memSeries. Chunk counts are bounded by the 3-byte field in HeadChunkRef, so cannot overflow uint32.

Should address the same problem as #14752, but with a different approach.

prombench (PS: @bwplotka made me aware in retrospect that prombench is currently found to be unreliable within the +-10% range):

-2.6% for /api/v1/query
-10% for /api/v1/query_range

Benchmark results (Intel Xeon Platinum 8280, 16 CPUs)

                                             │   main.txt   │          optimizations.txt          │
                                             │    sec/op    │    sec/op     vs base               │
MmapHeadChunks/series=1000/ready=10-16         374.1µ ±  3%    377.2µ ± 2%        ~ (p=0.485 n=6)
MmapHeadChunks/series=1000/ready=100-16        446.9µ ±  2%    441.8µ ± 2%        ~ (p=0.093 n=6)
MmapHeadChunks/series=1000/ready=1000-16       972.6µ ±  1%    988.7µ ± 3%        ~ (p=0.394 n=6)
MmapHeadChunks/series=10000/ready=100-16       921.5µ ±  3%    875.3µ ± 2%   -5.01% (p=0.002 n=6)
MmapHeadChunks/series=10000/ready=1000-16      1.619m ±  3%    1.495m ± 3%   -7.61% (p=0.002 n=6)
MmapHeadChunks/series=10000/ready=10000-16     7.594m ± 16%    7.485m ± 3%        ~ (p=0.394 n=6)
MmapHeadChunks/series=100000/ready=1000-16     7.405m ± 14%    3.425m ± 5%  -53.75% (p=0.002 n=6)
MmapHeadChunks/series=100000/ready=10000-16    19.14m ±  6%    12.22m ± 3%  -36.15% (p=0.002 n=6)
MmapHeadChunks/series=100000/ready=100000-16   97.23m ±  2%   100.07m ± 3%        ~ (p=0.065 n=6)
geomean                                        3.086m          2.664m       -13.67%

                                             │   main.txt    │         optimizations.txt          │
                                             │     B/op      │     B/op      vs base              │
MmapHeadChunks/series=1000/ready=10-16         2.687Ki ±  4%   2.677Ki ± 3%       ~ (p=0.621 n=6)
MmapHeadChunks/series=1000/ready=100-16        16.25Ki ±  4%   16.07Ki ± 1%       ~ (p=0.180 n=6)
MmapHeadChunks/series=1000/ready=1000-16       141.9Ki ±  1%   142.1Ki ± 1%       ~ (p=0.937 n=6)
MmapHeadChunks/series=10000/ready=100-16       16.94Ki ±  5%   16.68Ki ± 3%       ~ (p=0.180 n=6)
MmapHeadChunks/series=10000/ready=1000-16      144.3Ki ±  1%   142.2Ki ± 1%  -1.48% (p=0.041 n=6)
MmapHeadChunks/series=10000/ready=10000-16     1.382Mi ±  8%   1.387Mi ± 1%       ~ (p=0.461 n=6)
MmapHeadChunks/series=100000/ready=1000-16     172.3Ki ±  8%   157.8Ki ± 2%  -8.44% (p=0.002 n=6)
MmapHeadChunks/series=100000/ready=10000-16    1.342Mi ± 11%   1.363Mi ± 5%       ~ (p=0.290 n=6)
MmapHeadChunks/series=100000/ready=100000-16   12.99Mi ±  2%   12.99Mi ± 2%       ~ (p=0.171 n=6)
geomean                                        159.7Ki         157.7Ki       -1.25%

                                             │  main.txt   │          optimizations.txt          │
                                             │  allocs/op  │  allocs/op   vs base                │
MmapHeadChunks/series=1000/ready=10-16          11.00 ± 0%    11.00 ± 0%       ~ (p=1.000 n=6) ¹
MmapHeadChunks/series=1000/ready=100-16         105.0 ± 1%    105.0 ± 1%       ~ (p=1.000 n=6)
MmapHeadChunks/series=1000/ready=1000-16       1.035k ± 0%   1.035k ± 0%       ~ (p=1.000 n=6)
MmapHeadChunks/series=10000/ready=100-16        106.0 ± 1%    106.0 ± 1%       ~ (p=1.000 n=6)
MmapHeadChunks/series=10000/ready=1000-16      1.042k ± 0%   1.040k ± 0%  -0.19% (p=0.006 n=6)
MmapHeadChunks/series=10000/ready=10000-16     10.81k ± 0%   10.81k ± 0%       ~ (p=0.883 n=6)
MmapHeadChunks/series=100000/ready=1000-16     1.089k ± 0%   1.062k ± 0%  -2.39% (p=0.002 n=6)
MmapHeadChunks/series=100000/ready=10000-16    11.35k ± 1%   11.08k ± 1%  -2.45% (p=0.002 n=6)
MmapHeadChunks/series=100000/ready=100000-16   139.7k ± 2%   139.7k ± 2%       ~ (p=0.273 n=6)
geomean                                        1.105k        1.099k       -0.56%
¹ all samples are equal

At production-relevant scales (100k series, 1–10% ready), the optimized path is 36–54% faster with a 14% geomean improvement across all cases. When 100% of series are ready it's within noise of the full scan. Memory and allocations are essentially flat.

Which issue(s) does the PR fix:

Does this PR introduce a user-facing change?

[PERF] tsdb: Skip clean series during periodic head chunk mmap using cached head chunk count

aknuds1 · 2026-03-11T07:58:18Z

/prombench main

prombot · 2026-03-11T07:58:22Z

⏱️ Welcome to Prometheus Benchmarking Tool. ⏱️

Compared versions: PR-18272 and main

After the successful deployment (check status here), the benchmarking results can be viewed at:

Available Commands:

To restart benchmark: /prombench restart main
To stop benchmark: /prombench cancel
To print help: /prombench help

aknuds1 · 2026-03-11T16:26:46Z

/prombench start

prombot · 2026-03-11T16:26:49Z

Incorrect /prombench syntax; command requires one argument that matches (master|main|v[0-9]+\.[0-9]+\.[0-9]+\S*) regex.

Available Commands:

To start benchmark: /prombench <branch or git tag to compare with>
To restart benchmark: /prombench <branch or git tag to compare with>
To stop benchmark: /prombench cancel
To print help: /prombench help

Advanced Flags for start and restart Commands:

--bench.directory=<sub-directory of github.com/prometheus/test-infra/prombench
- See the details here, defaults to manifests/prombench.
--bench.version=<branch | @commit>
- See the details here, defaults to master.

Examples:

/prombench v3.0.0
/prombench v3.0.0 --bench.version=@aca1803ccf5d795eee4b0848707eab26d05965cc --bench.directory=manifests/prombench

aknuds1 · 2026-03-11T16:27:05Z

/prombench main

prombot · 2026-03-11T16:27:08Z

⏱️ Welcome to Prometheus Benchmarking Tool. ⏱️

Compared versions: PR-18272 and main

After the successful deployment (check status here), the benchmarking results can be viewed at:

Available Commands:

To restart benchmark: /prombench restart main
To stop benchmark: /prombench cancel
To print help: /prombench help

aknuds1 · 2026-03-12T05:43:10Z

/prombench main

prombot · 2026-03-12T05:43:13Z

⏱️ Welcome to Prometheus Benchmarking Tool. ⏱️

Compared versions: PR-18272 and main

After the successful deployment (check status here), the benchmarking results can be viewed at:

Available Commands:

To restart benchmark: /prombench restart main
To stop benchmark: /prombench cancel
To print help: /prombench help

aknuds1 · 2026-03-12T12:56:37Z

/prombench main

prombot · 2026-03-12T12:56:40Z

⏱️ Welcome to Prometheus Benchmarking Tool. ⏱️

Compared versions: PR-18272 and main

After the successful deployment (check status here), the benchmarking results can be viewed at:

Available Commands:

To restart benchmark: /prombench restart main
To stop benchmark: /prombench cancel
To print help: /prombench help

aknuds1 · 2026-03-13T07:25:25Z

/prombench cancel

The periodic mmapHeadChunks cycle previously scanned every series across all stripes, acquiring a per-series lock on each, even though typically >99% have nothing to mmap. Add a dirtyForMmap flag (sync/atomic.Bool, 4 bytes) to memSeries. When a new head chunk is cut during append, the flag is set atomically. The new mmapDirtyHeadChunks function iterates series with a lock-free Load() to skip clean ones — only dirty series acquire the series mutex. The existing full-scan mmapHeadChunks is kept for Head.Close(). sync/atomic.Bool (4 bytes, align 4) is used instead of go.uber.org/ atomic.Bool (8 bytes due to nocmp) to fit in existing struct padding without growing memSeries. Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

Also use "ready" term instead of "dirty". Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

Cache the head chunk count in an atomic Uint32 instead of a boolean flag. This enables future optimization of the O(n) memChunk.len() traversal and detection of pathologically long head chunk chains. Reset headChunkCount on all paths that remove head chunks: snapshot restore, WAL replay, truncateChunksBefore, and mmapHeadChunks. In mmapHeadChunks, set the count after mmapChunks based on actual state to handle the race where truncation clears headChunks between the lock-free fast-path check and acquiring the series lock. Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

Fix pre-existing comment errors in code blocks touched by recent changes: headChunks.next → headChunks.prev (the field is prev), mmapHeadChunks() → mmapChunks() (the method called on each series), remove redundant line with grammar error, and clarify when headChunkCount is updated. Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

Address reviewer feedback: - Move headChunkCount update into cutNewHeadChunk and the histogram counter-reset paths, where head chunks are actually created. This is simpler than updating in observeChunkCreated and naturally avoids inflating the count for OOO chunk creation (krajorama, bboreham). - Move headChunkCount reset into mmapChunks so it is next to where headChunks is manipulated (krajorama). - Move >= 2 comment from field definition to the check site (bboreham). - Set headChunkCount on snapshot restore path. - Add test verifying OOO inserts do not inflate headChunkCount. Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

- Count snapshot-restored chunks instead of hardcoding 1 (krajorama). - Use uint32 for truncation loop counter to avoid cast (krajorama). Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

krajorama

LGTM

- Remove headChunkCount stores from no-op paths in mmapChunks, to avoid masking bugs by silently correcting stale counts (bboreham). - Drop observeChunkCreated extraction, revert to inline metrics since the function no longer carries headChunkCount logic (bboreham). - Note that chunk counts are bounded by the 3-byte HeadChunkRef field and cannot overflow uint32 (bboreham). - Merge duplicated serialisation comments in mmapHeadChunks doc (bboreham). - Reword "unlink these from s.headChunks" to "Remove the tail of the list" (bboreham). Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

bboreham

Lgtm.

…8272) tsdb: Skip clean series during periodic head chunk mmap The periodic mmapHeadChunks cycle previously acquired a per-series lock on every series, even though typically >99% have nothing to mmap. This was identified as a CPU bottleneck in Grafana Mimir. Add a headChunkCount field (sync/atomic.Uint32) to memSeries that tracks the number of head chunks. It is incremented in cutNewHeadChunk and the histogram new-chunk paths, and reset by mmapChunks and truncateChunksBefore. mmapHeadChunks uses a lock-free Load to skip series with fewer than 2 head chunks, avoiding the per-series lock for clean series. sync/atomic.Uint32 (4 bytes) is used instead of go.uber.org/atomic (8 bytes) to fit in existing struct padding without growing memSeries. Chunk counts are bounded by the 3-byte field in HeadChunkRef, so cannot overflow uint32. Also fix pre-existing comment inaccuracies in the touched code: headChunks.next -> headChunks.prev, mmapHeadChunks() -> mmapChunks() in the doc comment, and a grammar error. --------- Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com> Signed-off-by: Raphael Bizos <r.bizos@criteo.com>

…➔ v3.12.0) (#730) This PR contains the following updates: | Package | Update | Change | |---|---|---| | [quay.io/prometheus/prometheus](https://github.com/prometheus/prometheus) | minor | `v3.11.3` → `v3.12.0` | --- ### Release Notes <details> <summary>prometheus/prometheus (quay.io/prometheus/prometheus)</summary> ### [`v3.12.0`](https://github.com/prometheus/prometheus/releases/tag/v3.12.0): 3.12.0 / 2026-05-28 [Compare Source](prometheus/prometheus@v3.11.3...v3.12.0) - \[SECURITY] Remote-write: Reject snappy-compressed requests whose declared decoded length exceeds the 32MB. Thanks to [@hibrian827](https://github.com/hibrian827) for reporting it. [#18642](prometheus/prometheus#18642) - \[SECURITY] STACKIT SD: Fix secrets being exposed in plaintext via `/-/config` endpoint. Thanks to [@August829](https://github.com/August829) and [@Phaxma](https://github.com/Phaxma) for reporting. GHSA-39j6-789q-qxvh [#18649](prometheus/prometheus#18649) - \[CHANGE] TSDB/Agent: Adds Start Timestamp field to all WAL Histogram samples in memory; used `st-storage` flag is enabled. [#18221](prometheus/prometheus#18221) - \[FEATURE] API: Add `/api/v1/status/self_metrics` endpoint returning the current state of the Prometheus server's own metrics about itself as JSON. [#18411](prometheus/prometheus#18411) - \[FEATURE] Discovery: Add DigitalOcean Managed Databases service discovery [#18287](prometheus/prometheus#18287) - \[FEATURE] Prometheus: Add support for the aix/ppc64 compilation target [#18321](prometheus/prometheus#18321) - \[FEATURE] Discovery: Add Outscale VM service discovery (`outscale_sd_configs`) for discovering scrape targets from the Outscale Cloud API. [#18139](prometheus/prometheus#18139) - \[FEATURE] PromQL: Emit a warning when `sort`, `sort_by_label` or `sort_by_label_desc` is used within range (matrix) queries, as these functions do not have effect in that context. [#18498](prometheus/prometheus#18498) - \[FEATURE] PromQL: Add `start()`, `end()`, `range()`, and `step()` experimental functions [#17877](prometheus/prometheus#17877) - \[FEATURE] PromQL: Update `resets()` function to consider start timestamp resets. Hidden behind `use-start-timestamps` feature flag. [#18627](prometheus/prometheus#18627) - \[FEATURE] Prometheus: Promote auto-reload-config as stable [#18620](prometheus/prometheus#18620) - \[FEATURE] TSDB/Agent: Add `CheckpointFromInMemorySeries` option to `agent.DB` that enables checkpoint based on in-memory series. [#17948](prometheus/prometheus#17948) - \[FEATURE] UI: Add a web interface for deleting time series and cleaning tombstones, accessible from the Status menu. [#18390](prometheus/prometheus#18390) - \[FEATURE] PromQL: Use start timestamps for `rate()`, `irate(), and `increase()`calculations, behind a feature flag`use-start-timestamps`. Doesn't work together with extended range selectors `anchored`and`smoothed\`. [#18344](prometheus/prometheus#18344) - \[FEATURE] Scrape: Added a feature flag `st-synthesis` which synthesizes unknown STs for scraped cumulative metrics. Useful when Remote Writing 2.0 with delta or Otel-based backends. [#18279](prometheus/prometheus#18279) - \[FEATURE] promqltest: support `@st` annotation in `load` blocks to specify per-sample start timestamps. [#18360](prometheus/prometheus#18360) - \[ENHANCEMENT] API: reject concurrent fgprof profiles. [#18651](prometheus/prometheus#18651) - \[ENHANCEMENT] AWS SD: Add optional `external_id` field to ECS/MSK/RDS/Elasticache. [#18579](prometheus/prometheus#18579) - \[ENHANCEMENT] AWS SD: Add optional `external_id` field. [#17171](prometheus/prometheus#17171) - \[ENHANCEMENT] Discovery: Propagate SD target updates faster by introducing dynamic backoff interval instead of static 5s interval for throttling. [#18187](prometheus/prometheus#18187) - \[ENHANCEMENT] Promtool: Add `--header` flag to `query instant` command, matching existing `query range` behaviour. [#18418](prometheus/prometheus#18418) - \[ENHANCEMENT]: AWS SD: Allows EC2 service discovery to discover IPv6 addresses to communicate with target endpoints. The private IPv4 address remains the default when both IPv4 and IPv6 addresses are present. [#16088](prometheus/prometheus#16088) - \[PERF] TSDB: Make head chunk lookup in range queries constant time instead of quadratic time [#18302](prometheus/prometheus#18302) - \[PERF] TSDB: Skip entire stripes in mmapHeadChunks when no series need mmapping, reducing CPU utilization significantly at production-relevant scales. [#18541](prometheus/prometheus#18541) - \[PERF] TSDB: Skip clean series during periodic head chunk mmap using cached head chunk count [#18272](prometheus/prometheus#18272) - \[PERF] PromQL: Address FloatHistogram.KahanAdd performance regression on Go 1.26. [#18568](prometheus/prometheus#18568) - \[BUGFIX] PromQL: Fix `info()` function incorrectly handling negated `__name__` matchers [#17932](prometheus/prometheus#17932) - \[BUGFIX] API: Return duration expressions in `/parse_ast`. [#18624](prometheus/prometheus#18624) - \[BUGFIX] API: correctly document formats accepted for duration query request parameters (step, timeout and lookback delta) in OpenAPI spec [#18305](prometheus/prometheus#18305) - \[BUGFIX] Scrape: AppenderV2 now tracks staleness even when OOO/duplicate series errors happen similar to AppenderV1 [#18567](prometheus/prometheus#18567) - \[BUGFIX] Config: Validate remote\_write queue\_config fields at load time to prevent runtime panic and silent misconfiguration. [#18209](prometheus/prometheus#18209) - \[BUGFIX] Discovery/Consul: Add `health_filter` for Health API filtering, fixing breakage when using Catalog-only fields like `ServiceTags` in `filter`. [#18479](prometheus/prometheus#18479) [#18499](prometheus/prometheus#18499) - \[BUGFIX] OTLP: limit decompressed body size for gzip-encoded OTLP write requests. [#18408](prometheus/prometheus#18408) - \[BUGFIX] PromQL: Fix `smoothed` rate/increase returning zero instead of no result when all data falls strictly after the query range. [#18523](prometheus/prometheus#18523) - \[BUGFIX] PromQL: Fix metric name not being dropped when last\_over\_time or first\_over\_time is applied to subqueries containing name-dropping functions like abs(). [#18409](prometheus/prometheus#18409) - \[BUGFIX] PromQL: Fix missing warning when mixing exponential and custom-bucket histograms in stats queries. [#18660](prometheus/prometheus#18660) - \[BUGFIX] PromQL: Fix parsing of `range()` keyword in duration expressions such as `foo[5m+range()]`. [#18623](prometheus/prometheus#18623) - \[BUGFIX] PromQL: Fix smoothed vector selector returning no results in binary operations when the `@` modifier is used. [#18531](prometheus/prometheus#18531) - \[BUGFIX] PromQL: Reject NaN, infinite, and out-of-range duration expressions instead of silently producing an out-of-range time.Duration. [#18639](prometheus/prometheus#18639) - \[BUGFIX] Scrape: Fix panic when scraping malformed native histograms. [#18414](prometheus/prometheus#18414) - \[BUGFIX] Scrape: fix panic when scraping a target exposing a summary with no quantiles via the protobuf format. [#18382](prometheus/prometheus#18382) - \[BUGFIX] Scrape: fix scrape failure log file occasionally not applied after a configuration reload. [#18421](prometheus/prometheus#18421) - \[BUGFIX] TSDB: Allow retention percentage with new data path. [#18628](prometheus/prometheus#18628) - \[BUGFIX] TSDB: Preserve decimal precision in percentage-based retention [#18374](prometheus/prometheus#18374) - \[BUGFIX] TSDB: fix prometheus\_tsdb\_head\_chunks going negative after WAL replay [#18401](prometheus/prometheus#18401) - \[BUGFIX] TSDB: panic with native histograms during query of overlapping chunks. [#18692](prometheus/prometheus#18692) - \[BUGFIX] Tracing: fix startup failure for insecure OTLP HTTP tracing [#18469](prometheus/prometheus#18469) - \[BUGFIX] UI: Escape label values offered by PromQL autocomplete. [#18658](prometheus/prometheus#18658) - \[BUGFIX] UI: Improve Y-axis tick label precision for graph values over small ranges. [#18682](prometheus/prometheus#18682) - \[BUGFIX] `prometheus_sd_refresh*` and `prometheus_sd_discovered_targets` metrics for specific scrape jobs are deleted when the scrape job is removed. [#17614](prometheus/prometheus#17614) - \[BUGFIX] Remote: fixed validation for received RW2 requests when parsing metadata unit symbols. This fixes a case when request would cause (recovered) handler panic. [#18641](prometheus/prometheus#18641) - \[BUGFIX] TSDB/Agent: fix race in agent appender where concurrent appends for the same label set could produce duplicate in-memory series and duplicate WAL records. [#18292](prometheus/prometheus#18292) - \[BUGFIX] Config: Update `--enable-feature` flag description and sort feature names. [#18487](prometheus/prometheus#18487) </details> --- ### Configuration 📅 **Schedule**: Branch creation - At any time (no schedule defined), Automerge - At any time (no schedule defined). 🚦 **Automerge**: Disabled by config. Please merge this manually once you are satisfied. ♻ **Rebasing**: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox. 🔕 **Ignore**: Close this PR and you won't be reminded about these updates again. --- - [ ] If you want to rebase/retry this PR, check this box --- This PR has been generated by [Renovate Bot](https://github.com/renovatebot/renovate).  Reviewed-on: https://git.erwanleboucher.dev/eleboucher/homelab/pulls/730

aknuds1 added component/tsdb kind/performance labels Mar 11, 2026

prombot added the prombench label Mar 11, 2026

aknuds1 force-pushed the arve/optimize-mmap-headchunks branch 2 times, most recently from 1136bf7 to 36dacdd Compare March 11, 2026 10:55

aknuds1 force-pushed the arve/optimize-mmap-headchunks branch 2 times, most recently from a019fba to 7a639ad Compare March 12, 2026 05:15

aknuds1 force-pushed the arve/optimize-mmap-headchunks branch 4 times, most recently from 5d1bcb0 to 653abf0 Compare March 12, 2026 11:39

aknuds1 marked this pull request as ready for review March 12, 2026 12:56

aknuds1 requested review from a team, bwplotka, codesome, jesusvazquez and krajorama as code owners March 12, 2026 12:56

aknuds1 marked this pull request as draft March 12, 2026 13:12

aknuds1 marked this pull request as ready for review March 12, 2026 13:23

aknuds1 force-pushed the arve/optimize-mmap-headchunks branch from 6fd8854 to 98c438b Compare April 14, 2026 07:26

aknuds1 requested a review from krajorama April 14, 2026 08:55

krajorama reviewed Apr 14, 2026

View reviewed changes

Comment thread tsdb/head.go Outdated

bboreham reviewed Apr 14, 2026

View reviewed changes

Comment thread tsdb/head.go Outdated

Comment thread tsdb/head_append.go Outdated

Comment thread tsdb/head_append.go Outdated

This was referenced Apr 14, 2026

Keep track of what series have new head chunks to avoid iterating over all timeseries every minute to mmap new chunks #14752

Draft

Allow mmap chunks during head compaction #14610

Open

aknuds1 added 7 commits April 14, 2026 15:20

Apply reviewer feedback

af2a32f

Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

Apply reviewer feedback

0f1b054

Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

Consolidate around optimized mmapHeadChunks

ddff905

Also use "ready" term instead of "dirty". Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

aknuds1 force-pushed the arve/optimize-mmap-headchunks branch from 205c896 to ccf6491 Compare April 14, 2026 13:28

aknuds1 requested a review from krajorama April 14, 2026 13:29

Improve doc comment

906c99d

Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

krajorama reviewed Apr 14, 2026

View reviewed changes

Comment thread tsdb/head_wal.go Outdated

Comment thread tsdb/head.go Outdated

tsdb: Apply reviewer feedback

cfaf871

- Count snapshot-restored chunks instead of hardcoding 1 (krajorama). - Use uint32 for truncation loop counter to avoid cast (krajorama). Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

aknuds1 requested a review from krajorama April 14, 2026 14:14

krajorama approved these changes Apr 14, 2026

View reviewed changes

bboreham reviewed Apr 14, 2026

View reviewed changes

Comment thread tsdb/head_append.go Outdated

Comment thread tsdb/head_append.go Outdated

Comment thread tsdb/head.go

Comment thread tsdb/head.go Outdated

Comment thread tsdb/head_append.go Outdated

bboreham approved these changes Apr 14, 2026

View reviewed changes

aknuds1 enabled auto-merge (squash) April 14, 2026 14:58

aknuds1 merged commit 98809e4 into prometheus:main Apr 14, 2026
39 checks passed

aknuds1 deleted the arve/optimize-mmap-headchunks branch April 14, 2026 15:11

aknuds1 mentioned this pull request Apr 19, 2026

tsdb: skip entire stripes in mmapHeadChunks via per-stripe ready count #18541

Merged

krajorama mentioned this pull request Apr 27, 2026

tsdb: complete TestHistogramCounterResetHeader for integer histograms… #18289

Open

Conversation

aknuds1 commented Mar 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Which issue(s) does the PR fix:

Does this PR introduce a user-facing change?

Uh oh!

aknuds1 commented Mar 11, 2026

Uh oh!

prombot commented Mar 11, 2026

Uh oh!

aknuds1 commented Mar 11, 2026

Uh oh!

prombot commented Mar 11, 2026

Uh oh!

aknuds1 commented Mar 11, 2026

Uh oh!

prombot commented Mar 11, 2026

Uh oh!

aknuds1 commented Mar 12, 2026

Uh oh!

prombot commented Mar 12, 2026

Uh oh!

aknuds1 commented Mar 12, 2026

Uh oh!

prombot commented Mar 12, 2026

Uh oh!

aknuds1 commented Mar 13, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

krajorama left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

bboreham left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

aknuds1 commented Mar 11, 2026 •

edited

Loading