Skip to content

Releases

v1.6.0

Choose a tag to compare

@justinhwang justinhwang released this 25 Sep 21:06
4fc419f

Backwards Incompatible Changes

These affect users who import M3 as a Go library. Deployed binaries, on-disk formats and wire formats are unchanged.

  • All: Metrics scopes now use github.com/uber-go/tally/v4; the github.com/m3db/prometheus_* forks are no longer used and the instrument package exposes a single upstream prometheus/client_golang registry type (#4423)
  • All: github.com/m3db/stackmurmur3 is replaced by github.com/twmb/murmur3 (bit-identical output, verified by a parity test) (#4420, #4422)
  • All: Minimum Go version is now 1.26 (#4410, #4416, #4425)
  • All: etcd dependency moved from 3.5.5 to 3.6.13; the embedded etcd config uses the renamed embed.Config URL fields (ListenPeerUrls, ListenClientUrls, AdvertisePeerUrls, AdvertiseClientUrls) (#4407, #4416)

Features

  • M3Aggregator: Support ingestion routing policies, with a routing policy filter in M3Msg and dynamic configuration of consumer service filters via topic updates (#4297, #4378, #4381, #4386, #4388, #4389, #4390)
  • M3Aggregator: TLS support for the aggregator server and client, including TLS handshake on connect and a local TLS development environment (#4283, #4376, #4377, #4379)
  • M3Aggregator: Add percentage filter for consumer writers and allow multiple filters per consumer service (#4223, #4247)
  • M3Aggregator: Add read timeout, custom DialContext/ContextDialer and bytes emission metrics to the aggregator client (#4131, #4290, #4300, #4315)
  • M3Aggregator: Add TimerBatchSizes distribution metric and write success/error latency histograms (#4320, #4321, #4384, #4385)
  • M3Aggregator: Local Docker development environment with debug mode and a consumer service (#4374, #4375, #4380)
  • M3Coordinator: Add IncludeTags to R2 rollup options (#4277)
  • M3Coordinator: Add config for forwarding a percentage of unique series via remote write, and sampled logging of write/fetch host errors (#4173, #4190)
  • M3Coordinator: Sort Graphite find results (#4163)
  • M3DB: Circuit breaker middleware for client host queue writes, tripping only on timeout errors (#4335, #4336, #4373, #4393)
  • M3DB: Sub-cluster placement algorithm with validation for partial sub-clusters (#4355, #4391)
  • M3DB: Support static topology config with more hosts than the replication factor (#4239, #4242)
  • M3DB: Add x/encoding.Cloner to shallow clone response iterators (#4271)
  • M3DB: Add metrics for time taken to connect to the cluster at startup (#4267)
  • M3DB: Add verify_ids tool (#4244)
  • M3DB: Support etcd prefix KV retrievals and watches (#4333)
  • M3Msg: Graceful close for message writers, abort on server close, and stop accepting new connections before disconnecting existing ones (#4328, #4329, #4330, #4387)
  • M3Msg: Consumer writer flush latency metrics (#4326)
  • All: Optionally log build information as JSON, and expose a Zap config from the log config (#4097, #4100)
  • All: Release linux and darwin arm64 binaries and multi-arch (amd64 and arm64) Docker images (#4442)

Bug Fixes

  • M3Aggregator: Fix colliding filters in MultiCharSequenceFilter by matching the most complete pattern rather than the first (#4188)
  • M3Aggregator: Followers can lead before the first flush is persisted (#4169)
  • M3Aggregator: Fix a memory allocation issue in ConsumerServiceWriterMetrics (#4382)
  • M3Aggregator: Close the aggregator client connection asynchronously (#4327)
  • M3Coordinator: Fix protobuf tag collision with Prometheus (#4256)
  • M3Coordinator: Fix inverted interpretation of the Prometheus remote write shadow sampling config (#4210)
  • M3Coordinator: Fix etcd config default value handling and only configure env configs when specified (#4130, #4216)
  • M3Query: Fix matching of disjunctions that include the empty string (#4258)
  • M3DB: Fix regexp LRU cache bug (#4250)
  • M3DB: Write datapoints that share timestamp, value and annotation but differ in time unit (#4401)
  • M3DB: Add index results from subsequent bootstrap runs (#4193)
  • M3DB: Fix strictly increasing order of the shard entry list for FetchBlocksMetadata pagination (#4298)
  • M3DB: Count write acks from paired leaving/initializing shards (#4191)
  • M3DB: Fix nil pointer panic on SyncCluster failure and the Get method of writeBatchPooledReqPool (#4246, #4253)
  • M3DB: Fix wrong error variable returned in two places, found by nilnesserr (#4430)
  • M3Msg: Remove data race in the producer (#4160)
  • All: Satisfy the yaml Marshaler interface on structs with custom unmarshalers (#4213)

Performance

  • M3Aggregator: Pool locked aggregations, fix an unintended allocation during element flushing, and clean up the aggregated metric writer and encoder (#4108, #4109, #4110, #4112)
  • M3Coordinator: Optimize m3.IsRollupID to not allocate and allow iterator reuse in ReverseMatch (#4304, #4305, #4307)
  • M3Msg: Message pooling cleanup, specialized message writer acker map, field alignment fix for RefCountedMessage, and parallel consumer writer flushes (#4104, #4105, #4113, #4331)
  • M3DB: Pass the shard set as part of tick instead of storing it in namespace readers (#4323)

v1.5.0

Choose a tag to compare

@robskillington robskillington released this 07 Apr 21:38
e7df2b9

Changelog

Features and Performance

  • M3DB: Optimize snapshotting by 30x or more which can speedup bootstrapping for workloads with large snapshots. (#4093)

Bug Fixes

  • M3DB: Fix multi-segment field iterator support of double underscore prefixed fields alphanumerically before __m3ninx_id. (#4095)
  • M3DB: Proactively GC Index entries belonging to a closed shard after cluster topology change. (#4094)

v1.4.2

Choose a tag to compare

@wesleyk wesleyk released this 19 Jan 19:13
68eea9a

Changelog

Bug Fixes

  • M3Coordinator: Guard against duplicate rollup tags in aggregation matching logic. (#3950)

Performance

  • M3DB: Improve index query performance by avoid cloning of roaring bitmap in conjunctionSearcher. (#3948)
  • M3DB: Improve bootstrapping memory profile by fixing entry leak in series ref resolver. (#3980)
  • M3DB: Provide better balanced shard placement by fixing unbalanced initial shard allocation. (#4020)

v1.4.1

Choose a tag to compare

@wesleyk wesleyk released this 20 Nov 18:35
95b3d60

Changelog

Bug Fixes

  • M3Coordinator: Do not Close singleton MessageProcessors when closing connections. This fixes a panic introduced that affects M3Coordinator -> M3Aggregator communication. (#3934)

v1.4.0

Choose a tag to compare

@wesleyk wesleyk released this 19 Nov 23:14
81a7a7d

Changelog

Features

  • M3Query: Add write endpoint support for M3-Map-Tags-JSON header in InfluxDB path (#3816)
  • M3Query: Add support for last_over_time in M3Query engine (#3884)
  • M3Aggregator: Add p75/p25 as aggregation options (#3867)

Bug Fixes

  • M3DB: Fix M3TSZ to be deterministic when encoding high precision values (3872)
  • M3DB: Gracefully handle reads including documents with stale index state (3905)

Performance

  • M3Aggregator: Rework close and remove persitFlushTimesEvery semantics in leader flushing in favour of always persisting shard flush times on a successful flush for optimized graceful failovers (#3890)
  • M3DB: Optimize filesetFiles function during bootstrapping for namespaces with long retentions to prevent CPU spikes (#3900)
  • M3DB: Avoid loading blocks in memory for namespaces with snapshots disabled during bootstrapping to reduce memory usage (#3919)

v1.3.0

Choose a tag to compare

@schallert schallert released this 10 Oct 12:38
4cd1b14

Changelog

Features

  • M3Coordinator: Add support for Prometheus Remote Write storage backend for sending aggregated and unaggregated metrics (#3742, #3768, #3783, #3791, #3814, #3777)
  • M3Coordinator: Add support for InfluxDB write endpoint GZip compression, setting timestamp precision and allowing an empty request body (#3373)
  • M3DB: Add SYSCTL_VM_MAX_MAP_COUNT env var for sysctl-setter sidecar allowing for custom VM max map count (#3689)

Bug Fixes

  • M3DB: Fix writes briefly degrading when creating a new namespace due to coarse lock acquisition (#3765)
  • M3DB: Fix compiled regexp DFA cache eviction on full bug that can lead to slow memory leak with large number of unique regexps (#3806)

Performance

  • M3Coordinator: Update default M3Msg retry initial backoff from 1s to 5s to reduces timeout and retries in large clusters (#3820)
  • M3DB: Fix performance of reverse index queries that cover huge time ranges (#3813)

v1.2.0

Choose a tag to compare

@schallert schallert released this 18 Sep 16:49
6b5c855

Changelog

Features

  • M3Query: Support Prometheus matchers with match[] URL parameters in label endpoints (#3180)
  • M3Query: Support Prometheus start and end time URL parameters for label and series metadata endpoints (#3214)
  • M3Query: Add Graphite functions and update functions with new arguments that were missing (#3048, #3367, #3370, #3145, #3149, #3142, #3469, #3484, #3545, #3576, #3582, #3583, #3521, #3602, #3641, #3644, #3648)
  • M3Query: Fix Graphite treatment of ** to allow to match an empty segment instead of one or more (#3366, #3593)
  • M3Query: Add M3-Limit-Max-Range header to optionally truncate time range of queries (#3538)
  • M3Coordinator: Add ability to use an exclude by rollup rule to rollup metrics without specific dimensions (#3318)
  • M3DB: Use better heuristics to cap the series and aggregate query limits that individual DB nodes apply for a query so in larger clusters the query can be clamped earlier (#3516, #3518, #3519, #3520, #3527)
  • M3DB: Add repair option full_sweep and ability to force a repair via API call (#3573, #3550)

Bug Fixes

  • M3DB: Fix aggregate series metadata query limits (#3112)
  • M3Coordinator: Make bad aggregated namespace headers return bad request status code instead of internal server error (#3070)
  • M3Coordinator: Propagate Require-Exhaustive parameter for aggregate series metadata queries (#3115)
  • M3Query: Add determinism to Graphite sort and reduce functions (#3164)

Performance

  • M3DB: Rearchitect index segments to compact and expire series on block rotation instead of build a new segment for new block (#3464)
  • M3DB: Add postings list cache for searches and repopulate during active block index segment compaction before segment made visible for queries (#3671)
  • M3DB: Avoid allocating index entry fields per series and read from backing mmap directly (#3050, #3062, #3057)
  • M3DB: Avoid allocating series IDs when read from disk (#3093)
  • M3DB: Improve speed of tag byte reuse from ID for tags by speeding up search (#3075)
  • M3DB: Improve speed of bootstrap by using StreamingReadMetadata API for reads from disk (#2938)
  • M3DB: Improve speed of bootstrap by using an asynchronously evaluated series resolver API that can be written to while bootstrapping reliably (#3316)
  • M3DB: Add limits for total series being read at any one time globally (#3141)
  • M3DB: Restrict the time a query can hold an index worker to help allow small queries to continue to execute while larger ones are paused and resumed (#3269)
  • M3DB: Use adaptive WriteBatch allocations to dynamically match workload throughput and batch sizes (#3429)
  • M3Coordinator: Improve rule matching speed by improving per element rule matching and disabling cache which puts locks in the hot path (#3080, #3083)
  • M3Query: Improve speed of M3TSZ decoding by using 64 bit operations (#2827)
  • M3Query: Improve speed of M3TSZ decoding by using int64 type xtime.UnixNano instead of time.Time (#3515)
  • M3Query: Improve speed of quorum reads by improving multi-replica iterator (#3512)

v1.1.0

Choose a tag to compare

@schallert schallert released this 03 Feb 16:58
468ab1c

Changelog

Features

  • M3Coordinator: Add /ready endpoint for readiness probe which checks current write/read consistency level achievability (#2976)
  • M3Coordinator: Add per endpoint status code response codes and latency metrics (#2880)
  • M3Coordinator: Add Graphite carbon ingest latency metrics (#3045)
  • M3Coordinator: Add Graphite carbon ingest rule matcher contains to compliment regexp for faster matching (#3046)
  • M3Coordinator: Return 504 errors on timeout to downstream M3DB nodes or other cross-region coordinators (#2886)
  • M3Coordinator: Validate placements when using the raw placement upsert endpoint unless force set is specified (#2922)
  • M3Query: Add Graphite powSeries function (#3038)
  • M3Query: Add Graphite support for ** with metric path selectors (#3020)
  • M3DB: Add ability to configure regexp DFA and FSA limits (#2926)
  • M3DB: Add Alibaba Cloud storage class Kubernetes manifest for disk provisioning in Aliyun (#2908)

Bug Fixes

  • M3Coordinator: Always set content type JSON for error responses (#2917)
  • M3Query: Fix invalid query resulting in 500 instead of 400 (#2910)
  • M3Query: Allow Graphite variadic functions to omit variadic args (#2882)

Performance

  • M3DB: Skip out of retention index segments during bootstrap (#2992)

Documentation

  • All: Add clustering getting started guides for both Kubernetes operator and binaries deployment (#2795)

v1.0.0-rc.2

Choose a tag to compare

@schallert schallert released this 19 Nov 10:32

Changelog

3746bf0 [dbnode] Add validation to ensure index claims is singleton (#2877)
a0a96a7 [dbnode] Default HostID and Discovery config for minimal config (#2876)
251dc3d [dbnode] Add claims for index segments volume index (#2846)
5b5c050 [dbnode] Remove namespaces from example config and integration tests (#2866)
a0a4c61 [dbnode] Resurrect flaky test skip (#2868)
47cde34 [aggregator] Fix checkCampaignStateLoop (#2867)
2d8dfe5 [dbnode] implement deletion method in namespace kvadmin service (#2861)

v1.0.0

Choose a tag to compare

@robskillington robskillington released this 16 Nov 16:36

Changelog

Overview

This release makes breaking changes to the APIs and configuration to provide a simpler experience both for setup and operating M3.

  • New M3 website.
  • New M3 documentation.
  • Simple M3DB configuration and guides.
  • M3DB hard limits limits for high resiliency under load.
  • Bootstrap rearchitecture, now able to boostrap hundreds of millions of recently written datapoints in minutes for reads on restart.
  • Continued focus on baseline performance release-over-release.

Features

  • M3DB: Namespace resolution and retention now configured dynamically via API and stored in etcd instead of being defined statically in M3Coordinator configuration.
message DatabaseCreateRequest {
  // ...

  // Optional aggregated namespace to create in 
  // addition to unaggregated namespace
  AggregatedNamespace aggregated_namespace = 8;
}
  • M3DB: Minimal configuration file with default settings looks like:
coordinator: {}
db: {}

and includes common settings such as global query limits.

Backwards Incompatible Changes

Configuration

  • M3DB: db.bootstrap.bootstrappers removed
  • M3DB: db.config nested under db.discovery.config (discovery can optionally accept different types of defaults instead of a custom config)
  • M3DB: db.fs config key renamed to db.filesystem
  • M3DB: db.writeNewSeriesLimitPerSecond config key nested under db.limits as db.limits.writeNewSeriesPerSecond
  • M3Coordinator: cluster.namespaces.storageMetricsType removed
  • M3Coordinator: tagOptions.tagOptions no longer supports legacy type
  • M3Query: limits.perQuery.maxComputedDatapoints removed
  • M3Query: limits.perQuery.maxFetchedDatapoints removed
  • M3Query: limits.global.maxFetchedDatapoints removed
  • M3Query: cache removed
  • M3Query: listenAddress changed to always be resolved as a string from config. Format changed from
listenAddress:
  config: "..."
  value: "..."

to

listenAddress: "..."

API

  • M3DB: /services/m3db/database/config/bootstrappers dynamic bootstappers endpoint removed
  • M3Coordinator: Removed deprecated URL /api/v1/namespace in favor of stable preferred URL /api/v1/services/m3db/namespace
  • M3Coordinator: Removed deprecated URL /api/v1/namespace/init in favor of stable preferred URL /api/v1/services/m3db/namespace/init
  • M3Coordinator: Removed deprecated URL /api/v1/namespace/unagg in favor of stable preferred URL /api/v1/services/m3db/namespace/unagg
  • M3Coordinator: Removed deprecated URL /api/v1/placement in favor of stable preferred URL /api/v1/services/m3db/placement
  • M3Coordinator: Removed deprecated URL /api/v1/placement/init in favor of stable preferred URL /api/v1/services/m3db/placement/init

Misc

  • M3Query: Concept of data point limit enforcers removed in favor of the other remaining query limits (e.g. max series). This also removed metrics cost_reporter_datapoints, cost_reporter_datapoints_counter, and cost_reporter_over_datapoints_limit.