[Bug][Dashboard] Engineering Throughput PR time panels count PRs without a project_pr_metrics row as 0 hours
Maintainer antworten meist innerhalb von 2 Tagen
Bewertung
- Schwierigkeit
- 3/5
- Geschätzter Aufwand
- Ein halber Tag
- Anfängerfreundlichkeit
- 22/100
- Issue-Typ
- Bug
- Klarheit
- Klar beschrieben
- Aktivitätsstatus
- Veraltet
- Tech-Stack
- grafana, mysql, postgresql, sql
- Bereich
- data-visualization, databases
Rechercherichtung
The affected queries live in the Engineering Throughput and Cycle Time dashboard JSON files, in both MySQL and PostgreSQL variants, including the team-view file. Start with the LEFT JOIN project_pr_metrics plus coalesce(..., 0) in the PR Cycle Time query, and check whether PRs without a metrics row are excluded. Done means those PRs no longer count as 0 h and the panels show No data when no row exists. PR #9226 is already open against this issue, so check it before starting.
Vom Indexierungsmodell aus dem Issue-Text verfasst.
Beschreibung
Search before asking
- I had searched in the issues and found no similar issues.
What happened
The PR time panels of the Engineering Throughput and Cycle Time dashboards treat a merged PR that has no
row in project_pr_metrics as a PR that took 0 hours. The PR should be left out of the average instead.
Affected panels (MySQL and PostgreSQL variants of each dashboard):
| Dashboard | Panels |
|---|---|
engineering-throughput-and-cycle-time.json |
PR Cycle Time, PR Coding Time, PR Pickup Time, PR Review Time, PR Deploy Time |
engineering-throughput-and-cycle-time-team-view.json |
Total Cycle Time(h), PR Coding Time(h), PR Pickup Time(h), PR Review Time(h), PR Deploy Time(h) |
All ten queries have the same shape (MySQL version of PR Cycle Time, trimmed):
with _prs as (
SELECT
pr.id,
pr.merged_date as pr_merged_date,
coalesce(prm.pr_cycle_time/60, 0) as cycle_time -- a missing row becomes 0
FROM pull_requests pr
left join project_pr_metrics prm on pr.id = prm.id
join project_mapping pm on pr.base_repo_id = pm.row_id and pm.table = 'repos'
WHERE $__timeFilter(pr.merged_date) and pm.project_name in (${project})
GROUP BY 1,2,3
)
SELECT ..., avg(cycle_time) FROM _prs GROUP BY ...
The set of PRs and the time axis come from pull_requests, while the metric values come from
project_pr_metrics. Because of LEFT JOIN + coalesce(..., 0), a merged PR without a project_pr_metrics
row enters avg() as 0 and pulls the average down:
pull_requests project_pr_metrics value used by the panel
───────────── ────────────────── ───────────────────────
PR A (merged) ──▶ pr_cycle_time = 14h ──▶ 14h
PR B (merged) ──▶ pr_cycle_time = 10h ──▶ 10h
PR C (merged) ──▶ (no row) ──▶ 0h ◀ should be excluded
avg = 8h (expected: 12h)
project_pr_metrics is written only by the DORA plugin (calculateChangeLeadTime in
backend/plugins/dora/tasks/change_lead_time_calculator.go). The most common way to hit this bug is a project
with DORA metrics disabled. In that case pull_requests has merged PRs, but project_pr_metrics has no rows
for them, so all five panels show a flat 0 h. It looks as if every PR was finished instantly, when in fact
no PR time data is available. The dashboard's "Data Sources Required" text does not mention DORA, so users
have no hint about the cause.
The bug depends only on whether a PR has a matching project_pr_metrics row. The reason the row is missing
does not matter.
What do you expect to happen
PRs without a project_pr_metrics row should be left out of these averages (removed from the denominator)
instead of being counted as 0.
If no PR in the selected range has a row (e.g. DORA is not enabled for the project), the panels should show
"No data" instead of 0 h.
How to reproduce
- Create a project with a GitHub or GitLab repository and keep DORA metrics disabled in the project settings.
- Collect data so that
pull_requestscontains merged PRs. - Open Engineering Throughput and Cycle Time (or its Team View) and select the project.
- PR Cycle / Coding / Pickup / Review / Deploy Time show 0 h in every period, even though there are merged PRs.
To count the merged PRs that these panels average in as 0:
SELECT count(*)
FROM pull_requests pr
JOIN project_mapping pm ON pr.base_repo_id = pm.row_id AND pm.`table` = 'repos'
LEFT JOIN project_pr_metrics prm ON pr.id = prm.id
WHERE pr.merged_date IS NOT NULL
AND pm.project_name = '<your project>'
AND prm.id IS NULL;
Anything else
Proposed fix (open to discussion)
- In the 20 affected queries (2 dashboards × 5 panels × MySQL/PostgreSQL), change
LEFT JOIN project_pr_metricsto an innerJOIN. - Keep the existing
coalesce(..., 0)on the per-row values. The comment above it says the goal is to
"make sure cycle_time equals the sum of the four metrics". A row can still have a NULL coding, pickup, review
or deploy time, and those values are still treated as 0. All five panels still average over the same set of
PRs (those that have a row), so this intent is preserved. The only change is that PRs with no row at all are
dropped. - When DORA is enabled, every merged PR in the project gets a row, so those projects see no change in values.
Only the "no row" case changes, from 0 h to excluded (or "No data").
Not affected (no change needed)
dora-details-lead-timefor-changes.jsonuses the sameLEFT JOIN+coalescepattern, but filters on
$__timeFilter(prm.pr_deployed_date), which already drops PRs without a row.DORA.json,dora-by-team.json,kiro-dora.json,kiro-credits-dora.json,
github-copilot-dora-correlation.jsonandmonorepo-subprojects.jsoneither inner-join or select directly
fromproject_pr_metrics, orAVG()the nullable column withoutcoalesce(NULLs are skipped).
Related
- #7398 reported PR Cycle Time showing 0 when data is incomplete. This issue is a different case: the PR has
noproject_pr_metricsrow at all.
Version
v1.0.3-beta15 (the same queries are still on main at 4ad2f66df)
Are you willing to submit PR?
- Yes I am willing to submit a PR!
Code of Conduct
- I agree to follow this project's Code of Conduct
- Vorherrschende Sprache
- Go
- Sterne
- 3.2k
- Forks
- 819
- Ø Merge
- 2 T. 22 Std.
- Gemergte PRs (30 T.)
- 43
Entwicklungsumgebung
Startet den Dev-Container des Projekts im Browser, mit Ihrem eigenen GitHub-Konto.
- Kein Dockerfile und keine Docker-Compose-Datei
- Hat eine Pull-Request-Vorlage
- Kein Beitragsleitfaden
Erste Schritte
- Lesen Sie das ganze Issue und danach den Beitragsleitfaden des Projekts.
- Schreiben Sie ins Issue, dass Sie es übernehmen — das erspart doppelte Arbeit.
- Forken Sie das Repository und arbeiten Sie in einem Branch.
- Öffnen Sie einen Pull Request, der die Issue-Nummer nennt.
Mehr aus apache/devlake
-
[Bug][Project] Invalid metric plugin options are accepted on project save and only fail at the next pipeline runEvtl. vergeben @chenwei791129 hat das vor 1 Tag übernommen. Offen
Schwierigkeit 3/5 Ein halber Tag Anfängerfreundlichkeit 22/100
Maintainer antworten meist innerhalb von 2 Tagen
-
[Bug][e2e] Several plugin e2e tests fail on a fresh database because they depend on tables created by other testsEvtl. vergeben @chenwei791129 hat das vor 2 Tagen übernommen. Offen
Schwierigkeit 4/5 3-5 Tage Anfängerfreundlichkeit 35/100
Maintainer antworten meist innerhalb von 2 Tagen
-
[Feature][dora] Allow filtering which pull requests count toward Lead Time for Changes by branch nameEvtl. vergeben @chenwei791129 hat das vor 1 Tag übernommen. Offen
Schwierigkeit 5/5 Über eine Woche Anfängerfreundlichkeit 15/100
Maintainer antworten meist innerhalb von 2 Tagen
-
Schwierigkeit 3/5 Ein halber Tag Anfängerfreundlichkeit 62/100
Maintainer antworten meist innerhalb von 2 Tagen
-
Schwierigkeit 3/5 1-2 Tage Anfängerfreundlichkeit 42/100
Maintainer antworten meist innerhalb von 2 Tagen
Ähnliche Issues
-
documentation hacktoberfest priority: low skill: good first issue status: ready for dev
Schwierigkeit 2/5 Ein halber Tag Anfängerfreundlichkeit 85/100
hyperledger-labs/fabric-smart-client#2142 ·
Maintainer antworten meist innerhalb von 1 Tag
-
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 78/100
numberly/vault-db-injector#107 ·
Maintainer antworten meist innerhalb von 1 Tag
-
area/import kind/bug
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 62/100
Maintainer antworten meist innerhalb von 1 Tag
-
[DOC] Wrong sg_egress_ports range in the Terraform Databricks Guide for AWS Hub & Spoke with Network FirewallEvtl. vergeben @akhil9tiet hat das heute übernommen. Offendocs
Schwierigkeit 1/5 Unter einer Stunde Anfängerfreundlichkeit 72/100
databricks/terraform-provider-databricks#6080 · 1 Kommentar ·
Maintainer antworten meist innerhalb von 3 Tagen
-
spqr-coordinator usage shows 'run --config', but 'run' is not a subcommand (unknown command "run")Offen
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 72/100
pg-sharding/spqr#3092 · 1 Kommentar ·
Maintainer antworten meist innerhalb von 1 Tag