Audit log
Dexaflow writes an entry to its audit log when someone changes what runs:
registers or promotes a DAG version, pauses or deletes a DAG, triggers, marks
or deletes a run, clears or marks a task, and changes a connection, a variable
or a pool. It also records sign-ins, API token changes, operator service API
calls, and the secret checks made on behalf of running tasks. Entries live in the audit_log table of the
Dexaflow database, one row per action, and belong to a tenant.
What is recorded
| Event | When |
|---|---|
dag.version.register | A DAG version is registered (dexaflow deploy, POST /api/v2/dags/{dag_id}/versions). |
dag.version.set_current | The current version of a DAG changes, by a deploy or a promotion. |
dag.pause, dag.unpause | A DAG is paused or unpaused, including a register that sets it. |
dag.delete | A DAG is deleted with its versions and runs (DELETE /api/v2/dags/{dag_id}?deregister=true). |
dag.history.clear | A DAG’s runs are deleted and the DAG kept (DELETE /api/v2/dags/{dag_id}, the UI’s delete). |
dagrun.delete | One run is deleted. |
dagrun.<run_type>.trigger | A run is triggered, for example dagrun.manual.trigger. |
dagrun.mark.<state> | A run’s state is set, for example dagrun.mark.failed or dagrun.mark.queued for a re-run. |
taskinstance.clear | A task instance is cleared, one entry per task. |
taskinstance.mark.<state> | A task instance’s state is set by a user. |
connection.create, connection.update, connection.delete | A connection is created, changed or deleted. A POST of an existing id is an update. |
variable.create, variable.update, variable.delete | A variable is created, changed or deleted. A POST of an existing key is an update. |
pool.create, pool.update, pool.delete | A pool is created, resized or changed, or deleted. |
secret.scope_warning | A task asked for a secret its DAG did not declare. |
secret.liveness_would_deny, secret.liveness_denied | A secret was asked for by a task instance that was no longer live, under auth.secret_liveness_mode observe or enforce (ADR 0055). |
api_token.create, api_token.revoke | A user creates or revokes an API token. |
api_token.retire_expired | An expired token is replaced by a new one. |
oidc.login.success, oidc.login.failure, oidc.tenant_pin_rejected, oidc.jit_provision | Single sign-on through OIDC, and the first sign-in that creates the user. |
oidc.cli_code, oidc.cli_consent, oidc.cli_device_verify | dexaflow login through the browser. |
issuer.login.success, issuer.login.failure, issuer.bearer.failure | Sign-in through a trusted issuer, and a bearer token from it that is refused. |
ci_oidc.exchange | A CI job exchanges its identity token for a Dexaflow token. |
auth.login | A password sign-in while OIDC is off, accepted or refused for bad credentials. |
auth.break_glass | A password sign-in while OIDC is on, allowed or refused. |
user.create | A user is created. |
user.deactivate, user.activate | An admin deactivates or reactivates a user through PATCH /api/v2/users/{user_id}. |
service.tenant.ensure, service.tenant.suspend, service.tenant.resume, service.user.ensure | A call to the operator service API. |
retention.purge | The retention janitor deleted audit entries. |
Entries whose event starts with dag., dagrun., taskinstance. or
secret. are recorded on a DAG and also show on that DAG’s tab.
Each entry records an action that was accepted, apart from the sign-in and
token entries, which also record refusals and carry an outcome. An entry is
written in the same transaction as the change for DAG registration,
promotion and pause, so one never exists without the other. The other entries
are written after the change and are best effort: if the write fails, the
server logs it and the change stands.
What an entry holds
| Field | What it is |
|---|---|
event_log_id | The entry’s id. |
when | When the action happened. |
event | The action, from the table above. |
owner | The email of the user who acted. Empty for actions with no user, such as the secret checks, a CI exchange or the retention janitor. |
dag_id | The DAG, for entries recorded on a DAG. |
run_id, task_id, try_number | The run and task the action was on, where it has them. |
extra | Everything else, as a JSON object: the version, image and spec hash of a register, the versions on both sides of a set_current, the outcome and reason of a sign-in, the id and name of a token, the rows and cutoff of a purge. A connection, variable or pool entry carries its name under connection, variable or pool, the names of the fields the request set under fields, and for a pool the slots (and previous_slots on a resize). |
Secret values, passwords, API tokens and their hashes are never written to the
audit log. A connection entry names the password or extra field when the
request set it, never its value, and a variable entry never carries the value. A token entry carries the token’s id and name; a secret entry
carries the kind of secret and the task, not its value.
Read it in the UI
The Audit Log screen, in the Browse menu, lists the entries of your tenant,
newest first. A DAG’s own Audit Log tab lists the entries recorded on that
DAG. Reading the audit log needs the read:audit_log permission, which only
the admin role has, so the screen, the tab and the endpoint are for tenant
admins; other roles do not see the menu entry.
Read it through the API
GET /api/v2/eventLogs returns the same entries in Airflow’s EventLog shape:
curl -s "$DEXAFLOW_URL/api/v2/eventLogs?dag_id=nightly&limit=100" \
-H "Authorization: Bearer $TOKEN"
dag_idkeeps the entries of one DAG.limitis 50 by default and at most 1000, orserver.max_page_limitwhen that is lower.offsetpages through the list. For a long history, follow the cursor of the response instead: each page past the first then costs the same as the first, whatever its depth.
Each tenant reads only its own entries. A sign-in event whose tenant cannot be
resolved, such as one that names a tenant that does not exist, is recorded in
the default tenant, so a refused sign-in is never dropped.
How long it is kept
By default the audit log is kept forever. retention.audit_log_days
(DEXAFLOW_RETENTION_AUDIT_LOG_DAYS) deletes entries older than that many
days, in every tenant. Each cycle that
deletes entries writes a retention.purge entry with the number of rows and
the cutoff, so the purge itself stays on record.
Data retention explains how to turn it on with a
dry run first, and the configuration reference
lists the other retention settings, such as the batch size.
If you need the log for longer than the database keeps it, export it on a
schedule through the API, or replicate the audit_log table, before the
retention window closes.
What is not recorded
- Reads: listing DAGs, opening logs, reading variables or connections.
- A password sign-in refused before the credential check: one over the per-address limit, or one with a malformed body.
- Actions of the scheduler and the agents, such as retries, reschedules and task outcomes: those are in the run’s history and the task logs.