Changelog
Features
- M3Ctl: Add M3 command line tool for calling APIs and using YAML files to apply commands (#2097)
- M3Coordinator: Add public API to write annotations (i.e. arbitrary bytes), next to datapoints for things like storing exemplars (#2022, #2029, #2031)
- M3Coordinator: Add support for mapping rules, allowing metrics to be stored at different resolutions based on their labels/tags (#2036)
- M3Coordinator: Add Graphite mapping rule support (#2060) (#2063)
- M3Coordinator: Add community contributed InfluxDB write endpoint (at /api/v1/influxdb/write) (#2083)
- M3Coordinator: Add headers to pass along with request to remote write forward targets (#2249)
- M3Coordinator: Add retry to remote write forward targets (#2299)
- M3Coordinator: Add in-place M3Msg topic consumer updates with a PUT request (#2186)
- M3Coordinator: Add ability to rewrite tags for Prometheus remote write requests using header (#2255)
- M3Coordinator: Add config for multi-process launcher and SO_REUSEPORT listen servers for non-container based multi-process scaling (#2292)
- M3Query: Add Prometheus engine to compliment Prometheus Remote Read, improves performance by skipping serialization/deserialization/network overhead between Prometheus and M3Query (#2343, #2369)
- M3Query: Add header to support enforcing all queries in request to implicitly always include a given label/tag matcher (#2053)
- M3Query: Return headers indicating incomplete results for cross-regional fanout queries when remote fails or hits a limit (#2053)
- M3Query: Refactor query server to allow for custom handlers (#2073)
- M3Query: Add remote read debug parameters to look at raw data for a PromQL query and/or get results as JSON (#2276)
- M3Query: Add warnings for Prometheus queries to Prometheus query JSON response (#2265)
- M3Query: Add ability to set default query timeout by config (#2226)
- M3Aggregator: Add M3Msg aggregator client for high throughput point to point clustered buffered delivery of metrics to aggregator (#2171)
- M3Aggregator: Add rollup rule support for metrics aggregated with pre-existing timestamps, such as Prometheus metrics (#2251)
- M3Aggregator: Add aggregator passthrough functionality for aggregation in a local region forwarding to a remote region for storage (#2235)
Performance
- M3DB: Improve RSS memory management with madvise resulting in flat RSS usage with a steady workload as time passes block-over-block (#2037)
- M3DB: Improve bootstrapping performance by allowing bootstrapping to be performed in a single pass, now possible for a lot of bootstraps to take just minutes depending on retention (#1989)
- M3DB: Use zero-copy references to index data instead of copy-on-read index data for each query, substantially improving query throughput and performance (#1839)
- M3DB: Further improve peer bootstrapping performance by using a document builder rather than raw memory segments (#2078)
- M3DB: Concurrent indexing when building segments for newly inserted metrics (#2146)
- M3DB: Decode ReadBits decompression improvements (#2197)
- M3DB: Remove implicit cloning of time ranges to reduce allocs (#2178)
- M3Query: Substantially improve temporal function performance (#2049)
- M3Query: Improve datapoint decompression speed (#2176, #2185, #2190)
- M3Query: Read bits uses an optimized byte reader (#2205)
- M3Coordinator: Ensure coordinator not grow M3Msg buffer if message over max size (#2207)
Bug Fixes
- M3Aggregator: Take last value by wall clock timestamp not arrival time to avoid late arrivals overwriting actual later occuring values (#2199)
- M3DB: Validate indexable metrics for valid utf-8 prior to insert, also includes a utility for earlier M3DB versions to remove non-utf8 index data (#2046)
- M3DB: Remove incorrect error log message for missing schema with default non-protobuf namespaces (#2013)
- M3DB: Fixed memory leak causing index blocks to remain in memory after flushing (#2037)
- M3DB: Fix long standing possibility of double RLock acqusition (#2128)
- M3DB: Remove loop in fileset writer when previous fileset encountered an error writing out index files (#2058)
- M3DB: Instead of encountering an error skip entries for unowned shards in commit log bootstrapper (#2145)
- M3DB: Fix to avoid returning error when missing writable bucket with a cold flush (#2188)
- M3DB: Set defaults and expose configuration of TChannel timeouts, this avoids idle connection growth (#2173)
- M3DB: Account for Neg/Pos Offsets when building per field roaring bitmap posting lists (#2213)
- M3DB: Fix to build flush errors (#2229, #2217)
- M3Coordinator: Respect env and zone headers for topic API endpoints (#2159)
- M3Coordinator: Add support for Graphite Grafana plugin /find POST requests (#2153)
- M3Coordinator: Use tag options specified in config with M3Msg ingester (#2212)
- M3Coordinator: Only honor default aggregation policies if not matched by mapping rule (#2203)
- M3Query: Fix namespace resolve debug log not being written with multiple namespaces (#2211)
- M3Query: Fix to temporal function regression leading to inconsistent results (#2231)
- M3Query: Fix edge cases with cross-zonal query fanout and add verify utility (#1993)
- M3Query: Fix issue with histogram grouping (#2247)
Documentation
- M3Aggregator: Add M3 aggregator Grafana dashboard (#2064)
- M3Coordinator: Add documentation to write to multiple clusters from a single coordinator (#2187)
- M3DB: Add documentation about estimating number of unique time series (#2062)
- M3DB: Update namespace configuration documentation to use simpler duration specific keys (#2045)
Misc
- All: Upgrade to Go 1.13 and switch dependency management to Go modules (#2221)
- All: Add gauge metrics to measure the number of active routines for any worker pool (#2061)
- All: Allow for ${ENV_VAR_NAME} expansion with YAML configuration files (#2033)
- All: Add a utility for comparing performance and correctness across different versions of M3DB, enabling diffing the perf of different versions (#2044)
- All: Upgrade etcd client library to 3.4.3 (#2101)
- All: Include key name in watch errors (#2138)
- Development: Add HA Prometheus lab setup for dev M3 docker compose deployment (#2206)
- Development: Temporarily disable kubeval validation to allow builds on go 1.12 (#2241)
- Development: Add comparator value ingester for replaying functions against given data sets (#2224)
- Development: Logging improvements (#2222,#2225)
- M3Aggregator: Add a datasource variable and reuse it in all the panels of the aggregator dashboard (#2182)
- M3DB: Add client bad request/internal error distinction for metrics and sampled logs (#2201)
- M3DB: Add latency metrics to remote reads (#2027)
- M3DB: Add metrics for async replication worker pool utilization (#2059)
- M3DB: Remove carbon debug flag and rely on log debug level for debugging Carbon/Graphite mapping rules (#2024)
- M3DB: Add metric for BootstrappedAndDurable (#2210)
- M3DB: Use madvdontneed=1 in DB nodes to get a more accurate view of memory usage (#2242)
- M3DB: Add trace spans for database bootstrap process helping to identify all remaining slow code paths (#2216)
- M3Coordinator: Add power user API to custom set placement goal state for cluster membership and shards (#2108)
- M3Coordinator: Delete M3 aggregator related etcd keys when aggregator placement deleted (#2133)
- M3Coordinator: Add metrics for remote aggregator client and downsampler (#2165)
- M3Coordinator: Add aggregator client maxBatchSize config for configuring buffer for data sent to aggregator (#2166)
- M3Query: Allow both GET and POST for query APIs (#2055)
- M3Query: Only build amd64 architecture for m3query releases (#2202)