mirror of
https://github.com/andwati/zitadel.git
synced 2026-11-20 20:08:46 +00:00
5ddee5081e2a9153dc44dd4ed6dae0f7e0d3bfe2
4
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
|
|
4bc484dac9 |
perf(eventstore): read projection events per event type (#12753)
# Which Problems Are Solved - A projection that subscribes to several event types gets slower the further it falls behind. - The handler reads its events with one query: `aggregate_type = ANY(...) AND event_type = ANY(...)`, ordered by position and limited to the bulk limit. - The `es_projection` index returns each (aggregate type, event type) range in position order, but Postgres cannot merge those ranges. - So each batch reads every matching event after the projection's position, sorts them all, and keeps only the bulk limit. - A projection that is N events behind therefore reads about N rows per batch. Catching up costs about N² in total, and once events arrive faster than such batches can process them, the projection does not recover. - Every query that triggers the projection waits for that catch-up. Examples are `GetSession` and the notification handlers. # How the Problems Are Solved - **New opt-in, `SearchQueryBuilder.ScanEventTypesSeparately()`.** It applies to event queries for a single instance, ascending and limited, whose sub-queries filter only by aggregate and event types. - For such a query, the SQL layer reads each distinct (aggregate type, event type) combination in its own `ORDER BY … LIMIT` subquery, using `UNNEST … CROSS JOIN LATERAL`. - It then merges the results with the same sort key and limit. - Each subquery is one `es_projection` range read in position order, so it stops at the limit. A batch reads at most `limit × combinations` rows, however far behind the projection is. - Every other query renders exactly as before. - **The projection handler** now passes its real combinations, one sub-query per aggregate type instead of the cross product of all types, and sets the opt-in. - **Results are identical to the single query.** - The sort key (`position, in_tx_order, instance_id, aggregate_type, aggregate_id, sequence`) is a total order, and the global top n is contained in the union of each combination's top n. - The cursor from #12703 and the open-transaction cap apply inside every subquery. - A test against a real Postgres compares both shapes from every cursor and every position, for several limits. The test events include positions shared by several transactions, whose `in_tx_order` restarts. # Additional Changes This is a separate commit and can be split into its own PR: - `cmd/defaults.yaml` sets `BulkLimit: 1000` for the `sessions` projection. - `projection_state_latency` gets histogram buckets up to one week. Until now, every projection more than 30 minutes behind was reported in the same top bucket. # Additional Context These numbers come from a local, CPU-pinned sandbox, for comparing builds only. - **Database:** Postgres 17 in a VM with 8 vCPUs and 16 GiB. The `Eventstore.Autovacuum` and `Projections` settings mirror a production-like setup. - **Starting state:** `sessions8` 4.8M events behind, restored byte-identically for every run. Batch query for `sessions8`, `EXPLAIN ANALYZE`, median of 5 runs, parallel workers at the default: | | before | after | | --- | ---: | ---: | | at the backlog, limit 1000 | 832 ms | 4.0 ms | | at the backlog, limit 200 | 832 ms | 1.0 ms | | at the head (idle poll) | 0.14 ms | 0.19 ms | | generic plan (`plan_cache_mode=force_generic_plan`), at the backlog | 1,042 ms | 4.0 ms | Draining `sessions8` from that backlog for 600 s, 3 runs per build (median, with the range in brackets): | | before | after | | --- | ---: | ---: | | events/s | 975 (932-1,032) | 7,811 (7,523-7,955) | | mean batch time, including writes | 1.03 s | 0.13 s | | still behind after 600 s | 4.2M | 17k-277k | A new index `(instance_id, position, in_tx_order, aggregate_type, event_type)` with the old query was evaluated as an alternative and rejected. Under a generic plan, Postgres chose it for projections whose events are rare and walked the whole event stream: 430 ms instead of 0.15 ms. This is the behaviour that led to dropping `es_instance_position` in #10626. - Follow-up for #12703 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5 <[email protected]> Co-authored-by: Copilot Autofix powered by AI <[email protected]> |
||
|
|
11dbb1b277 |
feat(logging): add streams (#11435)
# Which Problems Are Solved Streams allow differentiating logs produced by different components of Zitadel. # How the Problems Are Solved The `backend/v3/instrumentation/logging` package now exposes convenience function for setting and getting a logger from the context. As well as high-level functions to emit log records at various levels. When constructing a new logger a "stream" needs to be specified: - **runtime**: General runtime logs, such as startup and shutdown messages. Default for logs that do not belong to the other categories. - **request**: Logs for incoming API and HTTP requests. - **event_handler**: Logs for event handling in projections. - **queue**: Logs for the job queue processing. - **event_pusher**: Logs for event pushing to the database. Disabled by default, contains sensitive information. Each line from the returned logger contains a `stream` field as well as a `version` field with the current Zitadel version. ## Runtime config Streams can be enabled by passing an array of stream names in the runtime config. Because some log streams may contain sensitive data (especially events), it is now also possible to mask values by their key. # Additional Changes - Wrap `slogctx` in the `logging` package. (Except API error converter packages, because of import cycle) - Add some docs to `logging` package so other devs understand how to add logging to Zitadel - Add `logging.OnError` and `logging.WithError` helper functions with `Panic()` and `Fatal()` methods, to preserve current calls in the `cmd` packages. - Add instance context extractor. - Only output request details in the request info log. Request ID remains propagated through context. - Moved middleware functionality into protocol specific packages. - Removed setting of URI to context in metric middleware. There were ony setters and no getters. (Unused value) - Reuse a single statusWriter in the middleware package for middlewares that need to know the response status. # Additional Context - Closes #11333 - Closes #11331 - Partly #11330 --------- Co-authored-by: Silvan <[email protected]> |
||
|
|
eb22b58756 |
feat(telemetry): improved instrumentation for observability (#11159)
# Which Problems Are Solved Zitadel did not provide easy correlation between errors, logs, traces and metrics. The configuration for those instrumentations was also not consistent, with some supporting different exporters then others. Implementation and parsing of config was also spaghettified over multiple packages, with awkward parsing and inconsistent naming of options. # How the Problems Are Solved All telemetry is now merged under the name "instrumentation". Why? 1. We thought it was a good idea in the past to call the milestone exporter `Telemtry` in the runtime config. Calling this `TelemetryV2` looks weird. 2. Not everything is a meter and not everything is sent (tele...). 3. It's also [defined](https://opentelemetry.io/docs/concepts/instrumentation/) as such by the OTEL documentation. ## New features - Adds structured, context based logging with trace-ID awareness - Static log fields are added to the context, such as service and request path - Static log fields are injected in each logline emitted by the application - Structured logs can also be send to an otel exporter - Structured logs can be printed to StdErr in text and JSON format - Error sinks make sure every error is logged at the correct level: - Warnings for client side errors (HTTP 400 range, Invalid request etc) - Error for server side errors (Internal server errors) - Metrics can now also be send to a OTEL collector. (previously they could only be scraped from `/debug/metrics` with prometheus) ## Exporters This change adds all the exporters supported by OTEL upstream and some google specific exporters for our cloud deployment. - StdOut / StdErr: all instrumentations - OTEL gRPC / HTTP: all instrumentations - Google: all instrumentations except logging - Prometheus (pull-based): only metrics The exception is profiling, which only supports the google exporting due to lack of support by OTEL upstream. ## Configuration and structure - All instrumentation is moved into the new `backend/v3/instrumentation` package. It reuses configuration types, so both code and runtime configuration are easier to understand. - The `internal/telemetry` packages are removed. - Instrumentation is started with a single function and a proper shutdown function is now provided. - Legacy configuration is still parsed from the runtime config, as long as the new configuration is disabled. This allows backporting this feature to v4 without breaking existing configurations. # Additional Changes - Devcontainer: set `$PATH` variable so installed go binaries can be run individually, without NX. - NX: install GCI tool to fix imports # Additional Context - Closes https://github.com/zitadel/zitadel/issues/8408 - Closes https://github.com/zitadel/zitadel/issues/6664 - Backport to v4 |
||
|
|
c1535b7b49 |
feat: add prometheus metrics on projection handlers (#9561)
# Which Problems Are Solved With current provided telemetry it's difficult to predict when a projection handler is under increased load until it's too late and causes downstream issues. Importantly, projection updating is in the critical path for many login flows and increased latency there can result in system downtime for users. # How the Problems Are Solved This PR adds three new prometheus-style metrics: 1. **projection_events_processed** (_labels: projection, success_) - This metric gives us a counter of the number of events processed per projection update run and whether they we're processed without error. A high number of events being processed can let us know how busy a particular projection handler is. 2. **projection_handle_timer** _(labels: projection)_ - This is the time it takes to process a projection update given a batch of events - time to take the current_states lock, query for new events, reduce, update_the projection, and update current_states. 3. **projection_state_latency** _(labels: projection)_ - This is the time from the last event processed in the current_states table for a given projection. It tells us how old was the last event you processed? Or, how far behind are you running for this projection? Higher latencies could mean high load or stalled projection handling. # Additional Changes I also had to initialize the global otel metrics provider (`metrics.M`) in the `setup` step additionally to `start` since projection handlers are initialized at setup. The initialization checks if a metrics provider is already set (in case of `start-from-setup` or `start-from-init` to prevent overwriting, which causes the otel metrics provider to stop working. # Additional Context ## Example Dashboards   --------- Co-authored-by: Silvan <[email protected]> Co-authored-by: Livio Spring <[email protected]> |