[BanyanDB] Stream write callback reports success for events it did not store
维护者通常 1 天内回复
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 48/100
- Issue 类型
- 缺陷
- 描述清晰度
- 描述清楚
- 活跃度
- 活跃
- 技术栈
- go
- 领域
- backend, observability
调研方向
Start with banyand/stream/write_standalone.go, then read the error forwarding paths in banyand/queue/local.go and banyand/queue/sub/helpers.go and how batch statuses are applied in banyand/liaison/grpc/stream.go. Check the corresponding callbacks in banyand/measure and banyand/trace. Done means failures before any event is stored reach the caller, while partial and index-write failures do not cause stored events to be retried as failures.
由索引模型根据 Issue 内容生成。
描述
Summary
writeCallback.Rev (banyand/stream/write_standalone.go) logs each failed event and continues, and always returns an empty response. The caller gets STATUS_SUCCEED for elements that were not stored.
For example, for a missing group, handle fails in prepareElementsInGroup (cannot load tsdb for group). Rev logs cannot handle write event and returns no error.
Observed
With native self-stored logs (apache/skywalking-banyandb#1419) on a standalone node, a forced delete of _monitoring_log lost 3 events. native_log_written_total rose for them, and every native_log_dropped_total reason stayed 0.
Notes for a fix
- The error reply path already exists:
localBatchPublisher.Close(banyand/queue/local.go) and the data-node sub server (banyand/queue/sub/helpers.go) forward a*common.Error, asCheckHealthdoes for disk-full. - The liaison applies one status to every message it sent to a node in a batch (
banyand/liaison/grpc/stream.go). Reporting an error on a partial failure would mark stored elements as failed, and a retrying client would store them twice. Report an error only when nothing in the batch was stored, or report per element. - Index-write failures happen after the elements are stored, so they should stay logged only.
banyand/measureandbanyand/traceuse the same pattern.
Effect on #1419
#1419 does not change this callback. Until it is fixed, events lost while the log destination is missing are counted as written, so a test cannot observe that loss.
- 主要语言
- Java
- 星标
- 25k
- 派生
- 6.6k
- 平均合并
- 16 小时 35 分钟
- 30 天内合并 PR
- 32
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
apache/skywalking 的其他 Issue
-
[BanyanDB] Flaky test: TestExportRelay_TwoRelaysShareOneRate可能已有人在做 @hanahmily 于 1 天前认领。 未关闭database test
难度 3/5 1-2 天 新手友好度 66/100
apache/skywalking#14175 · 已指派 1 人 ·
维护者通常 1 天内回复
-
[Bug] BanyanDB trace block metadata format changed in 0.11 without a file format version bump可能已有人在做 @hanahmily 于 4 天前认领。 未关闭bug database
apache/skywalking#14169 · 已指派 1 人 ·
维护者通常 1 天内回复
-
chore CLI database
难度 5/5 一周以上 新手友好度 25/100
apache/skywalking#14168 · 1 条评论 ·
维护者通常 1 天内回复
-
[Feature] Canopy auth 2/2: SSO (OIDC/OAuth2), break-glass and the auth design doc可能已有人在做 @hanahmily 于 5 天前认领。 未关闭database feature
难度 5/5 一周以上 新手友好度 20/100
apache/skywalking#14162 ·
维护者通常 1 天内回复
-
database
难度 4/5 3-5 天 新手友好度 58/100
apache/skywalking#14157 ·
维护者通常 1 天内回复
查看 apache/skywalking 的全部 Issue
相似的 Issue
-
area: triage theme: undefined
难度 2/5 1-3 小时 新手友好度 85/100
jhipster/generator-jhipster#35895 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 78/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 85/100
oracle/graalvm-reachability-metadata#11659 ·
维护者通常 1 天内回复
-
Build failure report and inbox copy miss a proof split by colour codes, which the build log redacts未关闭bug follow-up
难度 2/5 1-3 小时 新手友好度 85/100
Sanne/incus-spawn#1864 ·
维护者通常 1 天内回复
-
难度 1/5 1 小时以内 新手友好度 85/100
openhab/openhab-addons#22135 ·
维护者通常 1 天内回复