Skip to main content
Video tutorial · 0:47

DORA metrics

Insights → DORA metrics measures delivery performance with the four DORA metrics, computed from the merges Merge Pilot recorded. Merge Pilot does not see deployments, so a merge into a production branch stands for a deployment, and pipeline failures after the merge stand for failed changes.

Opening it​

Insights in the side menu opens on DORA metrics. Every role sees it, for their own company. Platform admins see the company selected in the header; without one, the page says Select a company (Choose a company in the header to see DORA metrics.).

What is counted​

  • Production merges: merge requests of the company merged into their repository's default branch (as Merge Pilot last read it from the VCS), within the time window. The line at the top names the default branches: Based on merges into each repository's default branch (main, trunk). Merges into other branches are not counted, and a repository whose default branch is not known yet has no production merges (… (none recorded yet)). When production goes through more than one branch, the company sets Production branches for DORA (Company → General): branch names and patterns such as release/*, with or without Repository default branch. The line then names them (Based on merges into each repository's default branch and the company's production branches (main, release/*), or Based on production merges (release/*) without the default branch). Through the API, production branches can also be named for one request (see below).
  • Every repository: merge requests of watched repositories, of repositories under change management, and those imported from a watched repository's history are all counted.
  • Time window: 7d, 30d (the default) or 90d: the merges of the last 7, 30 or 90 days up to now.
  • Daily and weekly figures use days in UTC; weeks start on Monday.

Without any production merge in the window the page says No production merges in this period (Pick a longer time window, or check back after the next production merge.).

The four metrics​

Each metric has a card with its value and a performance badge (Elite, High, Medium, Low, or No data when there is nothing to measure), and a trend chart. The information sign on each card shows how the metric is approximated.

Deploy frequency​

Approximated by MRs merged into production branches (not actual deployment events)

  • Value: the number of production merges divided by the days of the window, shown per day with one decimal (0.4/day).
  • Chart (Deployment frequency (daily)): the merges of each day that had any.
BandDaily average
EliteMore than 1 a day
HighAt least once a week (1/7 a day)
MediumAt least once a month (1/30 a day)
LowLess

Lead time​

Time from opening the MR to its merge into production

  • Each merge's lead time: from when the merge request was opened on the VCS (or, when that is not known, when Merge Pilot first recorded it) to its merge.
  • Value: the median (P50) of the window's lead times. P75 and P90 are in the API.
  • Chart (Lead time P50 (weekly)): the median of each week, by the week of the merge.
BandP50
EliteLess than one day
HighLess than one week
MediumLess than 30 days
Low30 days or more

Change failure rate​

Percentage of production merges with post-merge pipeline failures

  • A failed merge is a production merge that has at least one pipeline recorded on its merge request that failed and started at or after the merge.
  • Value: failed merges divided by all production merges of the window, as a percentage with one decimal.
  • Chart (Change failure rate (weekly)): the rate of each week, by the week of the merge.
BandRate
Elite5% or less
HighOver 5%, up to 10%
MediumOver 10%, up to 15%
LowOver 15%

MTTR (time to restore)​

Time from a failed production merge to the next successful merge on the same branch

How incidents are found, for each repository and target branch, with its production merges in the window in merge order:

  1. A failed merge (as defined for the change failure rate) starts an incident.
  2. Failed merges in a row are one incident, timed from the first of them.
  3. The incident ends at the next merge into the same branch that is not failed. Its time to restore is from the first failed merge to that merge.
  4. An incident that no later merge in the window ended is unrecovered: it has no time to restore and is not in the percentiles.

Only merges inside the window are looked at: a failure before the window, or a recovery after it, is not seen.

  • Value: the median (P50) time to restore of the recovered incidents. P75 and P90 are in the API.
  • Percentiles use the nearest-rank method: the P-th percentile of N sorted times is the one at rank ⌈P × N⌉. With four incidents of 1, 2, 3 and 4 hours, P50 is 2 hours, P75 3 hours and P90 4 hours; a single incident is every percentile.
  • Chart (MTTR P50 (weekly)): the median of each week, by the week of the incident's first failed merge. With fewer than three recovered incidents in the window the chart shows Insufficient data (N incidents) instead.
BandP50
EliteLess than one hour
HighLess than one day
MediumLess than one week
LowOne week or more

Without a recovered incident the badge is No data and the card shows no value. The same holds for lead time and change failure rate without production merges.

How times are shown​

Lead time and MTTR are shown as <1m, minutes (42m), or hours and minutes (53h 10m), also beyond a day. A chart without any point says Not enough data for trend.

Risk insights​

A card of findings about risk, shown when at least one applies:

LineShown whenWhat it means
N% of deployments outside change windowsSome production merges were outside the change windowsThe share of the window's production merges whose risk assessment found them outside every change window. Merges without a risk assessment count in the total but never as outside
N deployments during change freezes (CFR: x vs y overall)Some production merges have a risk assessment that marks them as in a change freezeTheir number, the change failure rate among them, and the overall change failure rate
High-risk MRs: x failure rate vs y for low-riskBoth high-risk (or, without any, critical) and low-risk production merges existThe change failure rate of the high-risk (or critical) merges against that of the low-risk ones
Change rejection rate: x (N of M change approvals decided in this period were rejected)At least one change approval was rejected in the windowOf the company's change approvals decided in the window (approved or rejected, in Merge Pilot or in the tracker), the share that was rejected. It counts every change approval, whatever branch its merge request targets and whether it merged

Change windows and freezes are set on the company's Change calendar tab; risk levels come from Risk.

By project​

A table of the repositories with production merges in the window, most merges first, 50 at a time (Show more (N left)):

Column (label in the UI)Shows
ProjectThe project's name and its repository ID
DeploysIts production merges
Lead timeThe median lead time of its merges
CFRIts change failure rate
PerformanceThe change failure rate's band (Elite to Low). The other metrics do not count here

Benchmarks​

The bands come from the DORA / Accelerate State of DevOps research (Benchmarks based on DORA/Accelerate State of DevOps research). They are fixed: they cannot be changed per company. The API also returns them in words (benchmarks).

Refreshing and export​

The page is read when it opens and when the window changes; it is not refreshed on its own. The server keeps a result for up to 60 seconds per company, window and branch list. The page has no export.

API​

GET /api/v1/dora with:

ParameterValues
window7d, 30d (the default), 90d
branchesProduction branches, comma-separated (exact names). Left out: the company's Production branches for DORA, else each repository's default branch
company_idPlatform admins only (required for them; others always get their own company)

The answer holds:

  • deployment_frequency: total, daily_average, trend[] (date, count), by_project[] (repo_id, name, count), performance;
  • lead_time: p50, p75, p90 (milliseconds), trend[] (week_start, p50, count), by_project[] (repo_id, name, p50, count), performance;
  • change_failure_rate: rate (percent), total_deploys, failed_deploys, trend[] (week_start, rate, total, failed), by_project[] (repo_id, name, rate, total, failed), performance;
  • mttr: p50, p75, p90 (milliseconds), incidents (recovered), unrecovered_failures, insufficient_sample (fewer than three incidents), trend[] (week_start, p50, incidents), performance;
  • change_rejection_rate: rate (percent), decided (change approvals approved or rejected in the window), rejected;
  • risk_correlation: df_by_risk_level (production merges per risk level), cfr_by_risk_level, lead_time_by_risk_level (median, milliseconds), out_of_window_rate, blackout_deploy_count, in_blackout_cfr;
  • window, branches (the production branches: those named, or the repositories' default branches), default_branches (true when each repository's default branch is counted), patterns (the company's production branches beyond the default branches), benchmarks, definitions and generated_at.

performance is elite, high, medium, low or no_data. A platform admin without company_id gets 400.

Reading the metrics is not recorded in the audit log.

  • Dashboard: the cycle time breakdown of merged merge requests, phase by phase.
  • Monitoring report: what change management would have done with watched repositories.
  • Change calendar: merges, freezes and collisions day by day.
  • CI: which pipelines Merge Pilot records on a merge request.

Video tutorial

Short videos of this feature working on real GitLab and Jira projects. English audio, English and Turkish subtitles. All video tutorials →

DORA Metrics and AnalyticsYouTube ↗