发布日期: 2024-05-08
版本号: v2.52.0

Prometheus最新版本更新包含功能新增、性能优化及问题修复。新增功能包括：Kubernetes服务发现模块添加追踪API请求失败的指标，支持端点切片角色中注入节点与区域元数据标签；Azure服务发现与远程写入支持使用官方授权SDK；告警模块支持原生直方图模板；Linode服务发现新增IPv6范围探测与区域过滤。性能优化方面，PromQL提升正则匹配查询与聚合运算效率，TSDB模块改进数据块压缩策略（优先处理内存占用较高的头部数据），并优化多项底层性能。可观测性增强包括统一日志中删除序列的计数字段、记录WAL重放期间的分块加载耗时、完善信号终止处理日志等。问题修复涵盖PromQL直方图统计偏差、标签重复、负值桶标准差计算错误，修复抓取模块直方图模式配置异常、TSDB空头压缩计数错误、远程读取内存泄漏、API请求取消响应未终止等缺陷，同时修复promtool工具在无直方图数据时的崩溃问题。

更新内容 （中文）

  • [变更] TSDB：修复对超出保留周期数据块的谓词检查，使其包含恰好位于保留边界的数据块。#9633
  • [功能] Kubernetes服务发现：新增prometheus_sd_kubernetes_failures_total指标用于追踪Kubernetes API请求失败次数。#13554
  • [功能] Kubernetes服务发现：在使用endpointslice角色时添加节点(node)和区域(zone)元数据标签。#13935
  • [功能] Azure服务发现/远程写入：支持使用Azure授权SDK。#13099
  • [功能] 告警：支持原生直方图模板化。#13731
  • [功能] Linode服务发现：支持IPv6范围发现和区域过滤。#13774
  • [优化] PromQL：提升带正则表达式匹配器的查询性能。#13461
  • [优化] PromQL：提升聚合运算符使用时的性能。#13744
  • [优化] PromQL：验证label_join的目标标签。#13803
  • [优化] 抓取：单次抓取期间出现重复序列时递增prometheus_target_scrapes_sample_duplicate_timestamp_total指标。#12933
  • [优化] TSDB：多项性能改进。#13742 #13673 #13782
  • [优化] TSDB：当需要压缩head时暂停常规块压缩（优先处理head以降低内存消耗）。#13754
  • [优化] 可观测性：改进信号处理终止时的日志记录。#13772
  • [优化] 可观测性：所有删除序列的日志行统一使用"num_dropped"键。#13823
  • [优化] 可观测性：在WAL回放期间记录chunk快照和mmaped chunk回放耗时。#13838
  • [优化] 可观测性：记录是否从WBL创建块进行压缩。#13846
  • [修复] PromQL：修复查询直方图时样本数量统计不准确的问题。#13667
  • [修复] PromQL：修复当直方图包含负桶时的histogram_stddev和histogram_stdvar计算。#13852
  • [修复] PromQL：修复特定查询中可能出现的指标结果标签名和值重复问题。#13845
  • [修复] 抓取：修复抓取时原生直方图schema因子设置问题。#13846
  • [修复] TSDB：修复创建WAL检查点统计时直方图样本计数问题。#13776
  • [修复] TSDB：修复压缩空head的情况。#13755
  • [修复] TSDB：在WAL检查点中统计浮点直方图。#13844
  • [修复] 远程读取：修复因请求中断导致的内存泄漏问题。#13777
  • [修复] API：当/api/v1/series/请求被取消时停止构建响应。#13766
  • [修复] promtool：修复promtool tsdb analyze --extended在没有原生直方图时panic的问题。#13976

更新内容 （原始）

  • [CHANGE] TSDB: Fix the predicate checking for blocks which are beyond the retention period to include the ones right at the retention boundary. #9633
  • [FEATURE] Kubernetes SD: Add a new metric prometheus_sd_kubernetes_failures_total to track failed requests to Kubernetes API. #13554
  • [FEATURE] Kubernetes SD: Add node and zone metadata labels when using the endpointslice role. #13935
  • [FEATURE] Azure SD/Remote Write: Allow usage of Azure authorization SDK. #13099
  • [FEATURE] Alerting: Support native histogram templating. #13731
  • [FEATURE] Linode SD: Support IPv6 range discovery and region filtering. #13774
  • [ENHANCEMENT] PromQL: Performance improvements for queries with regex matchers. #13461
  • [ENHANCEMENT] PromQL: Performance improvements when using aggregation operators. #13744
  • [ENHANCEMENT] PromQL: Validate label_join destination label. #13803
  • [ENHANCEMENT] Scrape: Increment prometheus_target_scrapes_sample_duplicate_timestamp_total metric on duplicated series during one scrape. #12933
  • [ENHANCEMENT] TSDB: Many improvements in performance. #13742 #13673 #13782
  • [ENHANCEMENT] TSDB: Pause regular block compactions if the head needs to be compacted (prioritize head as it increases memory consumption). #13754
  • [ENHANCEMENT] Observability: Improved logging during signal handling termination. #13772
  • [ENHANCEMENT] Observability: All log lines for drop series use “num_dropped” key consistently. #13823
  • [ENHANCEMENT] Observability: Log chunk snapshot and mmaped chunk replay duration during WAL replay. #13838
  • [ENHANCEMENT] Observability: Log if the block is being created from WBL during compaction. #13846
  • [BUGFIX] PromQL: Fix inaccurate sample number statistic when querying histograms. #13667
  • [BUGFIX] PromQL: Fix histogram_stddev and histogram_stdvar for cases where the histogram has negative buckets. #13852
  • [BUGFIX] PromQL: Fix possible duplicated label name and values in a metric result for specific queries. #13845
  • [BUGFIX] Scrape: Fix setting native histogram schema factor during scrape. #13846
  • [BUGFIX] TSDB: Fix counting of histogram samples when creating WAL checkpoint stats. #13776
  • [BUGFIX] TSDB: Fix cases of compacting empty heads. #13755
  • [BUGFIX] TSDB: Count float histograms in WAL checkpoint. #13844
  • [BUGFIX] Remote Read: Fix memory leak due to broken requests. #13777
  • [BUGFIX] API: Stop building response for /api/v1/series/ when the API request was cancelled. #13766
  • [BUGFIX] promtool: Fix panic on promtool tsdb analyze --extended when no native histograms are present. #13976

下载链接