# 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]>
<!-- CURSOR_AGENT_PR_BODY_BEGIN -->
# Which Problems Are Solved
Handler/v2 projections (`users14` and every other event projection) can
overlook committed events under load:
- `eventstore.commands_to_events` stamped `position` with `NOW()`
(transaction start). A slow create TX could commit after a faster TX
that started later, so the projector stored a later position and never
saw the earlier one.
- Resume used `OFFSET` on a **filtered** row count. That could skip
`in_tx_order = 1` at the next position (for example `user.human.added`).
- Two transactions in the same microsecond both start at `in_tx_order =
1`. A cursor of `in_tx_order > last` alone would drop the other TX’s
first event.
This does not repair rows that are already missing from projections.
# How the Problems Are Solved
- Setup step 77 replaces `eventstore.commands_to_events` so `created_at
= statement_timestamp()` and `position = EXTRACT(EPOCH FROM
clock_timestamp())` at INSERT time. Initialise `08_events_table.sql`
matches that for new databases. The function is `VOLATILE` to match
`clock_timestamp()`.
- The projector still drains in-flight writers with `pg_advisory_lock` /
`pg_advisory_unlock` in one statement, then caps with `position <=
EXTRACT(EPOCH FROM now())` (projector TX start). The exclusive lock is
**not** held across the event `SELECT`. The cap must stay `now()`, not
`clock_timestamp()`, or later inserts become a high-water mark.
- Resume is a lexicographic `EventSortKey` on `(position, in_tx_order,
instance_id, aggregate_type, aggregate_id, sequence)` via
`AfterEventSortKey`. SQL `OFFSET` is no longer used for projection
catch-up.
- Handler and FieldHandler persist the last event’s `in_tx_order` in
`current_states.filter_offset` (existing column, new meaning). Step 77
backfills matching rows from `events2`; unmatched rows keep their
previous value.
# Additional Changes
- `eventstore.Event` exposes `InTxOrder() uint32` (BaseEvent, repository
Event, v3 event, v1 models).
- v3 push SELECT/RETURNING now includes `in_tx_order` so newly written
events carry a real ordinal.
- `WithMinPosition` stays an inclusive `PositionAtLeast` floor and does
not smash the sort-key cursor.
- `SearchQuery.PositionAfter(decimal)` is unchanged (used by
access-token lookup). Do not confuse it with
`SearchQueryBuilder.AfterEventSortKey`.
- `commands_to_events` in setup 77 and initialise 08 is `VOLATILE
PARALLEL SAFE`. Shipped 40/64/70 stay `STABLE` because they still stamp
with `NOW()`.
- Drop unused `last_owner TEXT` DECLARE from setup 77 and initialise
`08_events_table.sql` (`i INTEGER` remains for the enforce-owner loop).
Shipped 40/64/70 keep the leftover.
- This PR’s migration is numbered 77 because `main` already has setup 76
(`76_users14_login_equality_indexes`). First-loop order is s70, then
main’s 76, then this 77.
# Additional Context
**Reviewer map**
| Start here | Why |
| --- | --- |
| `cmd/setup/77.sql` + `cmd/setup/77.go` | Position stamp + atomic
function replace/backfill after shipped `s70` and main’s s76 |
| `internal/eventstore/event_sort_key.go` + `search_query.go` |
Canonical cursor and `AfterEventSortKey` |
| `internal/eventstore/handler/v2/{state.go,handler.go}` |
Persist/resume cursor; `eventQuery` |
| `internal/eventstore/repository/sql/{query.go,postgres.go}` | Tuple
`>` filter, 6-column `ORDER BY`, await cap |
| `internal/eventstore/v3/push.go` / `push.sql` | Scan `in_tx_order` on
write |
Mechanical / low-risk: handler and query test fixtures, Event interface
stubs (`event_base.go`, repository Event, mocks, v1 models).
**Do not look for** a `70.sql` change — shipped migrations stay as on
`main`. Setup 40/64 still use `NOW()`; step 77 overwrites the function
afterward. Fresh `FirstInstance` (runs before `s70`) can still stamp
`NOW()` until 77; runtime pushes after 77 use `clock_timestamp()`.
Main’s `cmd/setup/76.go` is the users14 login-equality indexes, not this
stamp.
**Rollout**
1. Run setup so 77 replaces `commands_to_events` and backfills
`filter_offset`.
2. Restart ZITADEL so projectors load the new resume path.
3. Existing missing `users14` rows are not repaired by this PR.
**Tests**
- `go test ./internal/eventstore/... ./cmd/setup/`
- Sort-key SQL fixtures in `query_sort_key_test.go` (no `OFFSET`;
aggregate-id queries keep 6-column `ORDER BY`)
- `cmd/setup/77_test.go` asserts the embed has no unused `last_owner`
and still uses `clock_timestamp()` / `VOLATILE PARALLEL SAFE`
Related: #8509, #10480 (remove OFFSET), #10560 (advisory locks), #10710
(stable ORDER BY).
Out of scope: repairing already-missing projection rows, holding the
exclusive lock during SELECT, rewriting historical setup 40/64 SQL.
<!-- CURSOR_AGENT_PR_BODY_END -->
<div><a
href="https://cursor.com/agents/bc-3da6eb51-11d4-4a8a-8ba0-e190f116df05?cursor_ref=pr_footer&cursor_cta=open_in_web"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://cursor.com/assets/images/open-in-web-dark.png"><source
media="(prefers-color-scheme: light)"
srcset="https://cursor.com/assets/images/open-in-web-light.png"><img
alt="Open in Web" width="114" height="28"
src="https://cursor.com/assets/images/open-in-web-dark.png"></picture></a> <a
href="https://cursor.com/background-agent?bcId=bc-3da6eb51-11d4-4a8a-8ba0-e190f116df05&cursor_ref=pr_footer&cursor_cta=open_in_cursor"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://cursor.com/assets/images/open-in-cursor-dark.png"><source
media="(prefers-color-scheme: light)"
srcset="https://cursor.com/assets/images/open-in-cursor-light.png"><img
alt="Open in Cursor" width="131" height="28"
src="https://cursor.com/assets/images/open-in-cursor-dark.png"></picture></a> </div>
---------
Co-authored-by: Cursor Agent <[email protected]>
Co-authored-by: Silvan <[email protected]>
This PR overhauls our event projection system to make it more robust and
prevent skipped events under high load. The core change replaces our
custom, transaction-based locking with standard PostgreSQL advisory
locks. We also introduce a worker pool to manage concurrency and prevent
database connection exhaustion.
### Key Changes
* **Advisory Locks for Projections:** Replaces exclusive row locks and
inspection of `pg_stat_activity` with PostgreSQL advisory locks for
managing projection state. This is a more reliable and standard approach
to distributed locking.
* **Simplified Await Logic:** Removes the complex logic for awaiting
open transactions, simplifying it to a more straightforward time-based
filtering of events.
* **Projection Worker Pool:** Implements a worker pool to limit
concurrent projection triggers, preventing connection exhaustion and
improving stability under load. A new `MaxParallelTriggers`
configuration option is introduced.
### Problem Solved
Under high throughput, a race condition could cause projections to miss
events from the eventstore. This led to inconsistent data in projection
tables (e.g., a user grant might be missing). This PR fixes the
underlying locking and concurrency issues to ensure all events are
processed reliably.
### How it Works
1. **Event Writing:** When writing events, a *shared* advisory lock is
taken. This signals that a write is in progress.
2. **Event Handling (Projections):**
* A projection worker attempts to acquire an *exclusive* advisory lock
for that specific projection. If the lock is already held, it means
another worker is on the job, so the current one backs off.
* Once the lock is acquired, the worker briefly acquires and releases
the same *shared* lock used by event writers. This acts as a barrier,
ensuring it waits for any in-flight writes to complete.
* Finally, it processes all events that occurred before its transaction
began.
### Additional Information
* ZITADEL no longer modifies the `application_name` PostgreSQL variable
during event writes.
* The lock on the `current_states` table is now `FOR NO KEY UPDATE`.
* Fixes https://github.com/zitadel/zitadel/issues/8509
---------
Co-authored-by: Copilot <[email protected]>
Co-authored-by: Tim Möhlmann <[email protected]>
# Eventstore fixes
- `event.Position` used float64 before which can lead to [precision
loss](https://github.com/golang/go/issues/47300). The type got replaced
by [a type without precision
loss](https://github.com/jackc/pgx-shopspring-decimal)
- the handler reported the wrong error if the current state was updated
and therefore took longer to retry failed events.
# Mirror fixes
- max age of auth requests can be configured to speed up copying data
from `auth.auth_requests` table. Auth requests last updated before the
set age will be ignored. Default is 1 month
- notification projections are skipped because notifications should be
sent by the source system. The projections are set to the latest
position
- ensure that mirror can be executed multiple times
---------
Co-authored-by: Livio Spring <[email protected]>
# Which Problems Are Solved
For truly event-based notification handler, we need to be able to filter
out events of aggregates which are already handled. For example when an
event like `notify.success` or `notify.failed` was created on an
aggregate, we no longer require events from that aggregate ID.
# How the Problems Are Solved
Extend the query builder to use a `NOT IN` clause which excludes
aggregate IDs when they have certain events for a certain aggregate
type. For optimization and proper index usages, certain filters are
inherited from the parent query, such as:
- Instance ID
- Instance IDs
- Position offset
This is a prettified query as used by the unit tests:
```sql
SELECT created_at, event_type, "sequence", "position", payload, creator, "owner", instance_id, aggregate_type, aggregate_id, revision
FROM eventstore.events2
WHERE instance_id = $1
AND aggregate_type = $2
AND event_type = $3
AND "position" > $4
AND aggregate_id NOT IN (
SELECT aggregate_id
FROM eventstore.events2
WHERE aggregate_type = $5
AND event_type = ANY($6)
AND instance_id = $7
AND "position" > $8
)
ORDER BY "position" DESC, in_tx_order DESC
LIMIT $9
```
I used this query to run it against the `oidc_session` aggregate looking
for added events, excluding aggregates where a token was revoked,
against a recent position. It fully used index scans:
<details>
```json
[
{
"Plan": {
"Node Type": "Index Scan",
"Parallel Aware": false,
"Async Capable": false,
"Scan Direction": "Forward",
"Index Name": "es_projection",
"Relation Name": "events2",
"Alias": "events2",
"Actual Rows": 2,
"Actual Loops": 1,
"Index Cond": "((instance_id = '286399006995644420'::text) AND (aggregate_type = 'oidc_session'::text) AND (event_type = 'oidc_session.added'::text) AND (\"position\" > 1731582100.784168))",
"Rows Removed by Index Recheck": 0,
"Filter": "(NOT (hashed SubPlan 1))",
"Rows Removed by Filter": 1,
"Plans": [
{
"Node Type": "Index Scan",
"Parent Relationship": "SubPlan",
"Subplan Name": "SubPlan 1",
"Parallel Aware": false,
"Async Capable": false,
"Scan Direction": "Forward",
"Index Name": "es_projection",
"Relation Name": "events2",
"Alias": "events2_1",
"Actual Rows": 1,
"Actual Loops": 1,
"Index Cond": "((instance_id = '286399006995644420'::text) AND (aggregate_type = 'oidc_session'::text) AND (event_type = 'oidc_session.access_token.revoked'::text) AND (\"position\" > 1731582100.784168))",
"Rows Removed by Index Recheck": 0
}
]
},
"Triggers": [
]
}
]
```
</details>
# Additional Changes
- None
# Additional Context
- Related to https://github.com/zitadel/zitadel/issues/8931
---------
Co-authored-by: adlerhurst <[email protected]>
# Which Problems Are Solved
We need a reliable way to lock events that are being processed as part
of a job queue. For example in the notification handlers.
# How the Problems Are Solved
Allow setting `FOR UPDATE [ NOWAIT | SKIP LOCKED ]` to the eventstore
query builder using an open transaction.
- NOWAIT returns an errors if the lock cannot be obtained
- SKIP LOCKED only returns row which are not locked.
- Default is to wait for the lock to be released.
# Additional Changes
- none
# Additional Context
- [Locking
docs](https://www.postgresql.org/docs/17/sql-select.html#SQL-FOR-UPDATE-SHARE)
- Related to https://github.com/zitadel/zitadel/issues/8931
# Which Problems Are Solved
Optimize the query that checks for terminated sessions in the access
token verifier. The verifier is used in auth middleware, userinfo and
introspection.
# How the Problems Are Solved
The previous implementation built a query for certain events and then
appended a single `PositionAfter` clause. This caused the postgreSQL
planner to use indexes only for the instance ID, aggregate IDs,
aggregate types and event types. Followed by an expensive sequential
scan for the position. This resulting in internal over-fetching of rows
before the final filter was applied.
Furthermore, the query was searching for events which are not always
applicable. For example, there was always a session ID search and if
there was a user ID, we would also search for a browser fingerprint in
event payload (expensive). Even if those argument string would be empty.
This PR changes:
1. Nest the position query, so that a full `instance_id, aggregate_id,
aggregate_type, event_type, "position"` index can be matched.
2. Redefine the `es_wm` index to include the `position` column.
3. Only search for events for the IDs that actually have a value. Do not
search (noop) if none of session ID, user ID or fingerpint ID are set.
New query plan:
# Additional Changes
- cleanup how we load multi-statement migrations and make that a bit
more reusable.
# Additional Context
- Related to https://github.com/zitadel/zitadel/issues/7639
# Which Problems Are Solved
Float64 which was used for the event.Position field is [not precise in
go and gets rounded](https://github.com/golang/go/issues/47300). This
can lead to unprecies position tracking of events and therefore
projections especially on cockcoachdb as the position used there is a
big number.
example of a unprecies position:
exact: 1725257931223002628
float64: 1725257931223002624.000000
# How the Problems Are Solved
The float64 was replaced by
[github.com/jackc/pgx-shopspring-decimal](https://github.com/jackc/pgx-shopspring-decimal).
# Additional Changes
Correct behaviour of makefile for load tests.
Rename `latestSequence`-queries to `latestPosition`
Even though this is a feature it's released as fix so that we can back port to earlier revisions.
As reported by multiple users startup of ZITADEL after leaded to downtime and worst case rollbacks to the previously deployed version.
The problem starts rising when there are too many events to process after the start of ZITADEL. The root cause are changes on projections (database tables) which must be recomputed. This PR solves this problem by adding a new step to the setup phase which prefills the projections. The step can be enabled by adding the `--init-projections`-flag to `setup`, `start-from-init` and `start-from-setup`. Setting this flag results in potentially longer duration of the setup phase but reduces the risk of the problems mentioned in the paragraph above.
* fix(eventstore): prevent allocation of filtered events
Directly reduce each event obtained from a sql.Rows scan,
so that we do not have to allocate all events in a slice.
* reinstate the mutex as RWMutex
* scan data directly
* add todos
* fix(writemodels): add reduce of parent
* test: remove comment
* update comments
---------
Co-authored-by: adlerhurst <[email protected]>
This implementation increases parallel write capabilities of the eventstore.
Please have a look at the technical advisories: [05](https://zitadel.com/docs/support/advisory/a10005) and [06](https://zitadel.com/docs/support/advisory/a10006).
The implementation of eventstore.push is rewritten and stored events are migrated to a new table `eventstore.events2`.
If you are using cockroach: make sure that the database user of ZITADEL has `VIEWACTIVITY` grant. This is used to query events.
* docs: update cockroachdb version to 22.2
* feat(adminAPI): ListEventTypes returns the list of event types ZITADEL implements
* feat(adminAPI): ListAggregateTypes returns the list of aggregate types ZITADEL implements
* feat(adminAPI): ListEvents allows `IAM_OWNERS` to search for events
## Note
This release requires a setup step to fully improve performance.
Be sure to start ZITADEL with an appropriate command (zitadel start-from-init / start-from-setup)
## Changes
- fix: only run projection scheduler on active instances
- fix: set default for concurrent instances of projections to 1 (for scheduling)
- fix: create more indexes on eventstore.events table
- fix: get current sequence for token check (improve reread performance)
* begin init checks for projections
* first projection checks
* debug notification providers with query fixes
* more projections and first index
* more projections
* more projections
* finish projections
* fix tests (remove db name)
* create tables in setup
* fix logging / error handling
* add tenant to views
* rename tenant to instance_id
* add instance_id to all projections
* add instance_id to all queries
* correct instance_id on projections
* add instance_id to failed_events
* use separate context for instance
* implement features projection
* implement features projection
* remove unique constraint from setup when migration failed
* add error to failed setup event
* add instance_id to primary keys
* fix IAM projection
* remove old migrations folder
* fix keysFromYAML test
Christian Jakob
