v0.4

Core:

• Optimise the SQL query for the Query page, by adding an index on the finish time and doing a case-insensitive search (@talex5 #215).
• Remove in-memory cache entries when no longer needed (@talex5 #235).
• Provide Analysis.quick_stat for faster stats (@talex5 #239). With large pipelines (above 20,000 boxes or so) collecting the Prometheus stats for pipeline states was getting slow.
• Improve error message if a cancel hook raises (@talex5 #214). Say which job raised the error.

Web UI:

• Add dependency on conf-graphviz (@dra27 #225).

Docker plugin:

• Add an architecture flag in the Docker builder to enable multiarch builds (including 32-bit) (@avsm #213).
• Add Current_docker.Raw.peek (@talex5 #226). This can be used to check the latest version of an image without pulling it.
• Add Docker Compose support to the Docker plugin (@avsm #228).
• Log in before creating manifests (@talex5 #232). Otherwise, we may hit the anonymous use limit.

Git plugin:

• Use --init with git submodule update --init --recursive (@talex5 #224). Otherwise, it's not really recursive.

GitHub plugin:

• Use the term allowlist for permitted usernames and make it case-insensitive (@avsm #210, @Julow #211).
• Handle larger numbers of app installations (@talex5 #219). GitHub sends them in batches, and we previously only used the first batch.

Examples:

• Replace custom logging in examples with Prometheus defaults (@talex5 #241).

Build fixes and cleanup:

• Adapt to the Alcotest 1.2 interface (@talex5 #212).
• Add missing dependency on ppx_deriving_yojson (@talex5 #209).
• Remove unused Option.map (@talex5 #216)
• Switch from "capnpc" to "capnp compile" (@MisterDA #233). capnpc is the old name, and isn't present on Windows.
• Switch to ocaml/opam as the base image for the Dockerfile (@avsm #230).
• Use Cmdliner.Term.term_result for correct error handling (@talex5 #241).

v0.3

The main changes are:

• "latching" of a previous pipeline stage output while building the new one. For example, while doing docker pull to check for updates to an image it will continue using the old one until the check finishes. Jobs with latched results are scheduled at a lower priority than jobs without any output yet.
• Many improvements to the web UI, including access controls, streaming of logs, improved querying, and custom extra pages.

Core:

• Allow user-defined pools (@talex5, #206). Add Pool.of_fn so that users can provide their own pools. This allows waiting for resources from an external service before marking a job as started.
• Add simple job priorities (@talex5, #187). Pools now have high and low priority queues. The cache puts jobs on the low priority queue if they have a latched result available.
• Allow jobs to latch previous output while rebuilding (@talex5, #178). If a rebuild is triggered by the schedule, do the rebuild in the background without changing the output to pending. This allows polling for changes without disrupting the rest of the pipeline. Such jobs are displayed with a gradient background (from the pending colour on the left to the latched output's colour on the right).
• Add Current_cache.Generic to control latching behaviour (@talex5, #179). This provides the full API. BUILDER and PUBLISHER are now special cases of GENERIC.
• Also latch failed results when cache entries expire (@talex5, #202).
• Don't clear latched result when auto-cancelling (@talex5, #195).
• Fix rebuilding of stale cache entries (@talex5, #169). If a build was already invalid when we loaded it from the database, we invalidated the key but didn't trigger an update properly.
• Fix Option.get crash when shutting down (@talex5, #205). If an error occurs on startup and a cache entry has expired then we set config to None to exit the engine, and the timer thread tries to read the config and crashes, hiding the real error.
• Fix scheduled builds for very fast jobs (@talex5, #185). We cleared the expiry timer when the job reported an active state. However, if the job was very quick (no async operations) we wouldn't see this state. Then we'd think the old timer was still in place and would be good to handle the new deadline too. Reported by Kate.
• Move Current_cache.output to new Instance submodule (@talex5, #201). The name "output" came from before the build and publish caches were unified and was confusing.
• Handle uncaught exceptions in monitor read functions (@talex5, #191).
• Add missing ?cwd to Current.Process.exec (@kit-ty-kate, #177).
• Delete directories with Lwt (@talex5, #192). Using Unix.unlink was causing high latencies in some cases.

Web UI:

• Improvements to query page (@talex5, #196).

  • Add toggles to query page to allow rebuilding multiple jobs at once.
  • Filter by job ID prefix in the query page. This makes it easy to restrict results to a particular day (or month, hour, etc).
  • Filter by operation type in query page.
  • Allow filtering by rebuild status in query page.
  • Show job queue time and run time in query page.
• Stream log files (@talex5, #165). If the job is still running, keep the connection open and stream the data as it arrives.
• Add Site and Context modules (@talex5, #170).
• Auto-generate links in nav-bar from routes (@talex5, #182). Expose routes in web API. This allows the user to provide whatever routes they like, or to use their own server.
• Update to new Routes API (@anuragsoni, #189 #190).

• Add a basic access control system (@talex5, #172).

  • Site now takes authn and has_role arguments.
  • authn (if given) is used to create a "Login" link in the navbar.
  • has_role is used to decide who can access which resource.
  • The GitHub plugin now provides an authentication backend that authenticates users with GitHub.
• Display the build history in the job page (@talex5, #166). Allows navigating to previous build results easily.

Diagram generation:

• Extend, rather than replace, the context in with_context (@talex5, #197).
• Truncate long tooltips in Graphviz diagrams. Graphviz rejects long messages with the error longer than 16384?.

Docker plugin:

• Return the RepoId after pushing a manifest (@talex5, #207).
• Add Raw module with low-level API (@talex5, #168). This is useful if you need to create your own custom components (for example, because you want to use a Current.t input to generate one of the fixed arguments such as ~run_args). See examples/docker_custom.ml for an example.

Git plugin:

• Add ?pool argument to Git.with_checkout (@talex5, #184). Git checkout operations can use a lot of CPU and IO, so allow users to supply a pool to limit the number of concurrent operations.
• Make Commit_id.pp_user_clone public (@talex5, #176). Also, Commit.id now returns a Commit_id.t. Use Commit.hash to get the hash.
• Fix clone instructions for GitHub PRs (@talex5, #164). GitHub only provides the refs/pull/NNN/head branches when fetching, not when cloning (with the default refspec). Detect this case and show an alternative command that the user can run to reproduce the operation.

GitHub plugin:

• Get all installation repositories, not just the first 30 (@talex5, #193). For now, as a precaution, ignore any installation that has configured all repositories to be tested (this is likely an accident).
• If listing repositories fails, retry after 30s (@talex5, #198).
• Add Installation.repositories ?include_archived option (@talex5, #203). Filtering archived repositories is now the default. There's not much point doing CI on them as you can't write the status result back.
• Add GitHub.Installation.account to get account name (@talex5, #204).
• Add Api.Anonymous.head_of (@talex5, #200). This allows monitoring the head of a public GitHub repository without needing an OAuth token. It's useful for the ocaml-ci-local command, allowing users to test the pipeline easily.
• Catch cohttp end-of-file exceptions (@talex5, #175).
• Remove unicorn from error string (@talex5, #183). When a GraphQL query returned a server error, we previously returned the body of the HTTP response as the error message. However, this includes a large picture of a unicorn, which is not suitable as an error string.

Slack plugin:

• Upgrade TLS library to cope with Slack's new TLS policy (@talex5, #174 #194).

Build:

• Update to dune 2 (@talex5, #171).
• Add missing alcotest dependency (@talex5, #163).

v0.2

The main new feature is that OCurrent now evaluates pipelines incrementally. This means that services with large pipelines (such as ocaml-ci, with around 10,000 stages) can decide what they need to do without using an excessive amount of CPU time.

• Replace changed promises with Engine.update (@talex5, #135)
• Evaluate pipelines incrementally (@talex5, #144)
• Avoid unnecessary re-evaluations in Current.list_map (@talex5, #148)
• Change the representation of terms to support incremental evaluation (@talex5, #150)

Documentation:

• Added Internals wiki page explaining how OCurrent works.
• Update README.md to link the API Docs (@shonfeder, #145)
• Fix typo in README ("about" -> "above") (@smolck, #100)

Docker plugin:

• Add support for run-arguments in docker plugin (@MagnusS, #96)
• Add ?build_args parameter to Current_docker build (@SquidDev, #102)
• Allow alternative Dockerfile filenames in Docker.build (@talex5, #111)
• Add Docker.pread to get container stdout as a string (@kit-ty-kate, #120)
• Make it easier to add custom Docker commands (@talex5, #130)

Git plugin:

• Git.fetch: don't take repository lock until the job starts (@talex5, #116)

GitHub plugin:

• Add Github.Api.head_of to track individual branches (@talex5, #114)
• Expose the low-level refs function to get access to branch names and PR numbers (@talex5, #123)
• Add Commit.uri to make links back to GitHub (@talex5, #131)
• Better error if setting a GitHub commit status fails (@talex5, #151)
• Only refresh the repository that generated the webhook event, not all of them (@talex5, #156)

Web UI:

• Use routes library to simplify routing (@talex5, #160)
• Add "/jobs" page with information about active jobs (@talex5, #125)
• Mark HTML pages as UTF-8 (@talex5, #128)

Dependencies:

• Update to new Alcotest API (@talex5, #121)
• Use mirage-crypto instead of nocrypto (@hannesm, #161)

Bug fixes:

• Fix analysis with hidden catch/state nodes (@talex5, #134)
• Allow cancelling a job while waiting for confirmation (@talex5 #137)
• Fix merging of error states for pairs (@talex5, #147)

Metrics:

• Report Prometheus metrics for pipeline stage states (@talex5, #115)
• Report metric for cache evaluations (@talex5, #143)

Build improvements:

• Make dune transitive dependencies explicit (@CraigFe, #97)
• Add OCaml-CI status badge to the README (@CraigFe, #107)

Other:

• Explicitly enumerate accepted --confirm values (@CraigFe, #108)
• Rename Input to Primitive and simplify the API (@talex5, #154)
• Call Lwt_main.yield while testing log patterns (@talex5, #155)
• Add Current.collapse to allow collapsing parts of diagrams (@talex5, #152)
• Add Current.with_context to help with diagram layout (@talex5, #159)

v0.1

Initial release.