[FF] `pipeline_analytics_siphon` -- Route PipelineAnalytics reads through siphon_p_ci_pipelines

Summary

This issue is to roll out the feature on production, that is currently behind the pipeline_analytics_siphon feature flag.

When enabled, the PipelineAnalytics GraphQL field (Project + Group) reads from the Siphon-replicated siphon_p_ci_pipelines table via ClickHouse::Finders::Ci::SiphonPipelinesFinder (argMax dedup pattern) instead of the ci_finished_pipelines_{hourly,daily} materialized views.

Owners

  • Most appropriate Slack channel to reach out to: #g_ci-platform
  • Best individual to reach out to: @narendran-kannan

Expectations

What are we expecting to happen?

  • /-/pipelines/charts (Project and Group) continues to render aggregate counts, per-status counts, and p50/p95/p99 duration statistics with parity to the MV path within a small expected delta (raw started_at filtering vs hour-bucketed started_at_bucket).
  • One bespoke sync pipeline (Ci::ClickHouse::DataIngestion::FinishedPipelinesSyncService) becomes redundant once the rollout is complete, freeing up Sidekiq capacity and removing a CSV-based ingestion path.

What can go wrong and how would we detect it?

  • Siphon query latency on large groups. siphon_p_ci_pipelines is row-level (no pre-aggregation), so 90+ day group queries read significantly more bytes than the daily MV. Detection: PipelineAnalytics request latency dashboards; GraphQL P95 latency for pipelineAnalytics field.
  • source enum translation bug. Siphon stores source as Nullable(Int64) enum; the finder translates the Ruby symbol via Ci::Pipeline.sources. A drift here would silently return zero rows for source-filtered queries. Detection: comparing aggregate counts between the two paths on a sample container.
  • Boundary semantic difference at hour boundaries. Siphon filters on raw started_at; MV uses hour-truncated started_at_bucket. This is intentional (siphon is more accurate) but customer-facing numbers may shift by a small fraction at the window boundaries.
  • ReplacingMergeTree dedup correctness. Finder uses argMax(_siphon_replicated_at) to pick the latest version of each row, and filters _siphon_deleted = false. A regression here would inflate counts. Detection: spot-check known soft-deleted pipelines.

Relevant dashboards:

  • GraphQL field latency: TBD
  • ClickHouse query latency: console.clickhouse.com

Rollout Steps

Note: Please make sure to run the chatops commands in the Slack channel that gets impacted by the command.

Rollout on non-production environments

  • Verify the MR with the feature flag is merged to master and has been deployed to non-production environments with /chatops gitlab run auto_deploy status <merge-commit-of-your-feature>
  • Deploy the feature flag at a percentage (recommended percentage: 50%) with /chatops gitlab run feature set pipeline_analytics_siphon 50 --actors --dev --pre --staging --staging-ref
  • Monitor that the error rates did not increase (repeat with a different percentage as necessary).
  • Enable the feature globally on non-production environments with /chatops gitlab run feature set pipeline_analytics_siphon true --dev --pre --staging --staging-ref
  • Verify that the feature works as expected. The best environment to validate the feature in is staging-canary as this is the first environment deployed to. Make sure you are configured to use canary.
  • If the feature flag causes end-to-end tests to fail, disable the feature flag on staging to avoid blocking deployments.
    • See #e2e-run-staging Slack channel and look for the following messages:
      • test kicked off: Feature flag pipeline_analytics_siphon has been set to true on **gstg**
      • test result: This pipeline was triggered due to toggling of pipeline_analytics_siphon feature flag

If you encounter end-to-end test failures and are unable to diagnose them, you may reach out to the #s_developer_experience Slack channel for assistance. Note that end-to-end test failures on staging-ref don't block deployments.

Before production rollout

  • If the change is significant and you wanted to announce in #whats-happening-at-gitlab, it best to do it before rollout to gitlab-org/gitlab-com.

Specific rollout on production

For visibility, all /chatops commands that target production must be executed in the #production Slack channel and cross-posted (with the command results) to the responsible team's Slack channel.

The flag uses a container actor (Project or Group), so the most natural enablement granularity is by group.

  • Ensure that the feature MRs have been deployed to both production and canary with /chatops gitlab run auto_deploy status <merge-commit-of-your-feature>
  • Enable for gitlab-org/gitlab and dogfood: /chatops gitlab run feature set --project=gitlab-org/gitlab pipeline_analytics_siphon true
  • Verify on https://gitlab.com/gitlab-org/gitlab/-/pipelines/charts that the numbers match what the MV path would return for the same window.
  • Expand to gitlab-org: /chatops gitlab run feature set --group=gitlab-org pipeline_analytics_siphon true
  • Expand to gitlab-com: /chatops gitlab run feature set --group=gitlab-com pipeline_analytics_siphon true

Preparation before global rollout

  • Set a milestone to this rollout issue to signal for enabling and removing the feature flag when it is stable.
  • Check if the feature flag change needs to be accompanied with a change management issue. Cross link the issue here if it does.
  • Ensure that you or a representative in development can be available for at least 2 hours after feature flag updates in production. If a different developer will be covering, or an exception is needed, please inform the oncall SRE by using the @sre-oncall Slack alias.
  • Ensure that documentation exists for the feature, and the version history text has been updated.
  • Ensure that any breaking changes have been announced following the release post process to ensure GitLab customers are aware.
  • Notify the #support_gitlab-com Slack channel and your team channel (more guidance when this is necessary in the dev docs).
  • If this flag is or may be queried by external API consumers (for example, IDE extensions, Duo CLI, or CI integrations), follow the external API consumer guidance and ensure a fail-open mechanism is in place before the rollout milestone is finalised.

Global rollout on production

For visibility, all /chatops commands that target production must be executed in the #production Slack channel and cross-posted (with the command results) to the responsible team's Slack channel.

Release the feature

After the feature has been deemed stable, the clean up should be done as soon as possible to permanently enable the feature and reduce complexity in the codebase.

You can either create a follow-up issue for Feature Flag Cleanup or use the checklist below in this same issue.

  • Create a merge request to remove the pipeline_analytics_siphon feature flag. Ask for review/approval/merge as usual. The MR should include the following changes:
    • Remove all references to the feature flag from the codebase.
    • Remove the YAML definitions for the feature from the repository.
    • Drop the dual-path clickhouse_model branch in CollectPipelineAnalyticsServiceBase.
    • Remove the per-path let overrides and the [true, false].each loops in:
      • spec/services/ci/collect_aggregate_pipeline_analytics_service_spec.rb
      • spec/services/ci/collect_time_series_pipeline_analytics_service_spec.rb
      • spec/requests/api/graphql/project/project_pipeline_analytics_spec.rb
      • spec/requests/api/graphql/group/group_pipeline_analytics_spec.rb
    • Drop the ci_finished_pipelines_{hourly,daily} models (or schedule their removal in a follow-up).
    • File a follow-up to retire Ci::ClickHouse::DataIngestion::FinishedPipelinesSyncService and the two ci_finished_pipelines_sync_*_workers ops flags.
  • Ensure that the cleanup MR has been included in the release package. If the merge request was deployed before the monthly release was tagged, the feature can be officially announced in a release blog post: /chatops gitlab run release check <merge-request-url> <milestone>
  • Close the feature issue to indicate the feature will be released in the current milestone.
  • Once the cleanup MR has been deployed to production, clean up the feature flag from all environments by running these chatops command in #production channel: /chatops gitlab run feature delete pipeline_analytics_siphon --dev --pre --staging --staging-ref --production
  • Close this rollout issue.

Rollback Steps

  • This feature can be disabled on production by running the following Chatops command:
/chatops gitlab run feature set pipeline_analytics_siphon false
  • Disable the feature flag on non-production environments:
/chatops gitlab run feature set pipeline_analytics_siphon false --dev --pre --staging --staging-ref
  • Delete feature flag from all environments:
/chatops gitlab run feature delete pipeline_analytics_siphon --dev --pre --staging --staging-ref --production
Edited by Narendran