Repository navigation
docs(tempo): weekly docs triage 2026-10-04 – 2026-10-12 — 4 items #8087
Copy link
Copy link
Open
Labels
type/docsImprovements or additions to documentationImprovements or additions to documentation
Description
Created by the docs-workforce automation (AI-assisted docs triage) from recently merged PRs in grafana/tempo. Verify each item against the code before acting.
What needs docs
- Docs update needed — #8056:
querier.partition_ring.read_lookback_periodis only in generatedmanifest.md. Add it todocs/sources/tempo/configuration/_index.mdunder querierpartition_ring, and clarify inactive-partition read eligibility indocs/sources/tempo/reference-tempo-architecture/partition-ring.md. - Docs update needed — #8068: TraceQL metrics
target_bytes_per_jobdefault (600MiB) is documented, but block-builderwal.ingestion_time_range_slackdefault is now 2h in code whileconfiguration/_index.mdstill omits it underblock_builder.wal(shared WAL section still says 2m). - Docs update needed — #8062:
tempo-cli parquet rewritegained--row-group-size-bytes,--bloom-fp, and--bloom-shard-size-bytes(with inherit/default behavior). The command is not documented indocs/sources/tempo/operations/tempo_cli.md. - Docs update needed — #8058: Poller skips blocks with undecodable
meta.jsoninstead of failing the tenant poll, and emitstempodb_blocklist_corrupt_block_meta_total. Neither the behavior nor the metric appears indocs/sources/tempo/operations/monitor/polling.md.
Classification table
| PR | Title | Classification | Notes |
|---|---|---|---|
| #8080 | chore: upgrade Tempo to Go 1.27.2 | No docs required | Dependency / toolchain bump |
| #8076 | Skip predicate evaluation while seeking parquet iterators to a target row | No docs required | Internal Parquet iterator performance; no user-visible change |
| #8069 | chore(tempo-cli): count distinct tag names in benchmark | No docs required | Internal benchmark accounting |
| #8068 | Update defaults: block-builder slack 2h, TraceQL metrics job size 600MiB | Docs update needed | Metrics default updated in _index.md; block-builder slack default 2h missing from _index.md |
| #8067 | Allow to pin memcached servers | Docs present | pin_servers documented in configuration/_index.md Cache section with proxy-pool guidance |
| #8065 | distributor: add trace and attribute size histograms | Docs present | Documented in configuration, manage-trace-ingestion, and OOM troubleshooting pages |
| #8064 | Remove vParquet3 block format | Docs present | parquet.md, upgrade.md, migrate-to-3.md, tempo_cli.md, block-format.md, dedicated_columns.md updated |
| #8063 | re-resolve memcached servers on open CB | No docs required | Automatic refresh on circuit-breaker open; no new config |
| #8062 | tempo-cli: support modifying params in parquet rewrite | Docs update needed | New CLI flags; parquet rewrite absent from tempo_cli.md |
| #8060 | metrics-generator: cache recent values in the per-label limiter | No docs required | Internal performance cache |
| #8059 | feat(query-frontend): report inspected bytes as jobs complete | Docs present | query-io-and-timestamp-distance.md updated for incremental emission |
| #8058 | poller: skip blocks with corrupt meta instead of failing the tenant poll | Docs update needed | Behavior + tempodb_blocklist_corrupt_block_meta_total not in polling.md |
| #8056 | bugfix: Add a lookback time for querying inactive live-stores | Docs update needed | New read_lookback_period; only generated manifest updated |
| #8053 | feat(frontend): add inflight jobs gauge | No docs required | New metric only; no docs page names inflight jobs |
| #8049 | fix conflicting merge of PRs | No docs required | Internal merge conflict fix |
| #8047 | Fix block-builder attaching partially decoded traces | Docs present | Quote: block-builder.md "organizes the consumed spans by tenant" — fix restores that behavior (also removes crash) |
| #8046 | distributor: correct Kafka record split-size accounting | Docs present | Quote: configuration/_index.md producer_max_record_size_bytes — "An incoming write request larger than this size is split into multiple Kafka records." |
| #8036 | refactor(benchmark): move block profiling into pkg/benchmark/profile | No docs required | Internal refactor; no behavior change |
| #8015 | Reduce noise from idle backend scheduler | No docs required | Log noise only |
| #7967 | fix: shut down query-frontend and querier gracefully | No docs required | Hang / SIGKILL fix on shutdown |
| #7955 | redaction: allow the existence check attr != nil | Docs present | tempo_cli.md updated for attr != nil existence check |
| #7526 | vParquet5 better trace organization and compression | No docs required | Internal sort/compression; dedicated-columns batching fix covered by existing vParquet5 limits docs |
Screenshot inventory
None. No UI changes in this window.
Context
- Source repo:
grafana/tempo - Window:
2026-10-04–2026-10-12 - Docs root:
docs/sources/tempo/ - Orchestrator filtered out: 1 bot PR(s), 2 CI/test/deps/build-only PR(s)
run: dw-2026-10-12-b1156e8e
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
type/docsImprovements or additions to documentationImprovements or additions to documentation
Activity