Pre-existing Issue Alert Rules and Metric Alert Rules silently stop firing after upgrading to 26.7.0+ — never migrated to the Workflow Engine

[muneeba-mughal]

Self-Hosted Version

26.7.1

CPU Architecture

x86_64

Docker Version

N/A

Docker Compose Version

N/A

Machine Specification

• My system meets the minimum system requirements of Sentry

Installation Type

Upgrade from 26.5.0 to 26.7.1

Steps to Reproduce

1. Run a self-hosted instance predating the Workflow Engine, with existing Issue Alert Rules and/or Metric Alert Rules on various projects.
2. Upgrade to 26.7.0+, which enables the Workflow Engine (feat: enable workflow engine #4364).
3. Don't re-create or re-edit any pre-existing alert rules after the upgrade.
4. Check the Alerts UI for a project with an untouched pre-existing rule, and check whether that project has a default Error Monitor detector.

Expected Result

Pre-existing Issue Alert Rules and Metric Alert Rules keep firing after the upgrade — either auto-migrated to the new Workflow/Detector model, or still working via the old model as a fallback.

Actual Result

Rule and AlertRule objects are never migrated to the new Workflow/Detector model unless a human creates or edits that specific rule again after the upgrade. No automatic retroactive migration exists:

• The project-creation signal (workflow_engine/receivers/project_detectors.py) only fires for new projects.
• The one backfill migration touching existing data (workflow_engine/migrations/0094_backfill_issue_stream_detector_workflows.py) only operates on projects that already have an error detector — it can't create a first one.
• That migration and the equivalent for metric alerts are marked is_post_deployment = True, meant to be triggered manually post-deploy per Sentry's own convention. Self-hosted's install/upgrade tooling has no such trigger anywhere.

Impact on our instance (~430 active projects, ~1,300 active alert rules):

• 52 of 1,169 active Issue Alert Rules, across 24 projects, silently unmigrated and dead.
• 5 of 143 active Metric Alert Rules likewise unmigrated and dead.
• 19 of 432 active projects had zero default Error Monitor — blocking even new alert creation, since the UI only attaches alerts to an existing monitor and Error Monitors can't be created manually.

All silent — no error, no log, just an empty alert list.

Workaround: manually invoked IssueAlertMigrator/dual_write_alert_rule per affected rule via sentry django shell, gated on a "not already migrated" check (neither function is idempotent on its own). Worked cleanly for all 57 rules.

Suggested fix: document this in the upgrade guide, or ship an idempotent management command that backfills pre-existing Rule/AlertRule objects post-upgrade. Happy to help shape/test a fix.

Related (different symptoms, same 26.7.0 rollout): #4385, #4105, #4110, #4341, #4044

Event ID

No response

[smares]

Possibly related to my problem as well: #4424

[muneeba-mughal]

@smares your traceback in #4424 shows a Detector.DoesNotExist for the Issue Stream detector when sending a test notification. That looks like the same root cause as this issue: the project's default detectors were never backfilled after the workflow engine went live.

I looked deeply into this and found a workaround until it's fixed upstream, so it might help here too. If you have sentry django shell access, this should fix it for an affected project:

from sentry.workflow_engine.defaults.detectors import ensure_default_detectors
from sentry.models.project import Project

project = Project.objects.get(organization__slug="<your-org-slug>", slug="<your-project-slug>")
ensure_default_detectors(project)

This creates the missing Error Monitor and Issue Stream detectors for that project. It only creates what's missing, so you can call it again without creating duplicates.

If you also have pre-existing alert rules that stopped firing entirely, rather than just the test notification failing, those need an additional step. Each unmigrated Rule or AlertRule has to be migrated individually through IssueAlertMigrator or dual_write_alert_rule, which is what the rest of this issue covers. Happy to share more detail if that's useful.

[smares]

Thanks! The weird thing is that my old alerts still fire and we get Mails, but the alerts are nowhere to be found in the UI. 😅