github.com/travis-ci/worker

Added

• backend/docker: Allow disabling the CPU and RAM allocations by setting the config options to 0 (this was possible in previous versions, but was not documented or supported until now)

Changed

• backend/gce: Send job ID and repository slug to image selector to help in log correlation
• sentry: Only send fatal and panic levels to Sentry, with an option for sending errors as well (--sentry-hook-errors)
• image/api-selector: Send os:osx instead of the language when querying for an image matching an osx_image flag

Fixed

• amqp-job: Send correct received_at and started_at timestamps to hub in the case of the job finishing before the received or started event is sent

Added

• backend/docker: Tests added by @jacobgreenleaf :heart_eyes_cat:

Changed

• utils/package-*, bin/travis-worker-install: Rework and integration into base CI suite.
• Switch to using gvt for dependency management.
• amqp-job: Send all known timestamps during state updates.
• backend: Set defaults for all StartAttributes fields, which are also exposed via CLI.

Removed

• backend/bluebox

Fixed

• backend/docker: Use correct HostConfig when creating container
• image/api-selector: Set is_default=true when queries consist of a single search dimension.

Added

• image/api:
  • Add image selection query with dist + group + language
  • Add last-ditch image selection query with is_default=true
• log-writer: Write folded worker info summary

Changed

• utils/pkg: Official releases built with go 1.5.1.
• vendor/*: Updated all vendored dependencies
• utils/lintall: Set 1m deadline for all linters
• backend/jupiterbrain: switch to env image selector

Fixed

• backend/gce: Removed wonky instance group feature
• step/run-script: Do not requeue if max log length exceeded

Changed

• utils/pkg: updated upstart config to copy/run executable as /var/tmp/run/$UPSTART_JOB, allowing for multiple worker instances per host.

Added

• backend/gce:
  • Configurable image selector, defaulting to legacy selection method for backward compatibility.
  • Support for reading account JSON from filename or JSON blob.
  • Optionally add all instances to configurable instance group.
• image/*: New image selection abstraction with env-based and api-based implementations.

Changed

• vendor/*: Upgraded all vendored dependencies to latest.
• utils/pkg:
  • Official releases built with go 1.5.
  • Packagecloud script altered to only use ruby stdlib dependencies, removing the need for bundler.
• backend/gce: Lots of additional test coverage.
• backend/*: Introduction of Setup func for deferring mutative actions needed for runtime.
• config: Addition of Unset method on ProviderConfig

Fixed

• processor: Fix graceful shutdown by using tryClose on shutdown channel.

Added

• backend/gce: Add auto implode, which will cause a VM to automatically shut down after a hard timeout.

Fixed

• logger: Make the processor= field in the logs not be empty anymore
• sentry: Stringify the err field sent to Sentry, since it's usually parsed as a struct, making it just {} in Sentry.

Added

• backend/local

Changed

• backend/jupiterbrain: Add exponential backoff on all HTTP requests
• sentry: Include stack trace in logs sent to Sentry
• step/generate-script: Add exponential backoff to script generation

Fixed

• backend/gce: Fix a bug causing VMs for a build language ending in symbols (such as C++) to error while booting
• log-writer: Fix a race condition causing the log writer to be closed before the logs were fully flushed.
• log-writer: Minimize locking in the internals of the log writer, making deadlocks less likely.
• processor: Fix graceful and forceful shutdown when there are still build jobs waiting.

Added

• backend/gce
• backend/jupiterbrain: Per-image boot time and count metrics
• step/upload-script: Add a timeout for the script upload (currently 1 minute)

Changed

• step/upload-script: Treat connection errors as recoverable errors, and requeue the job

Fixed

• backend/jupiterbrain: Fix a goroutine/memory leak where SSH connections for cancelled jobs wouldn't get cleaned up
• logger: Don't print the job UUID if it's blank
• processor: Fix a panic that would sometimes happen on graceful shutdown

Changed

• config: Use the server hostname by default if no Librato source is given
• version: Only print the basename of the binary when showing version

Fixed

• step/run-script: Print the log timeout and not the hard timeout in the log timeout error message [GH-49]

Added

• Runtime pool size management: Send SIGTTIN and SIGTTOU signals to increase and decrease the pool size during runtime [GH-42]
• Report runtime memory metrics, including GC pause times and rates, and goroutine count [GH-45]
• Add more log messages so that all error messages are caught in some way

Changed

• Many smaller internal changes to remove all lint errors

Added

• backend/bluebox: (See #32)
• main: Lifecycle hooks (See #33)
• config: The log timeout can be set in the configuration
• config: The log timeout and hard timeout can be set per-job in the payload from AMQP (See #34)

Removed

• backend/saucelabs: (See #36)

Added

• backend/docker: Several new configuration settings:
  • CPUS: Number of CPUs available to each container (default is 2)
  • MEMORY: Amount of RAM available to each container (default is 4GiB)
  • CMD: Command to run when starting the container (default is /sbin/init)
• backend/jupiter-brain: New configuration setting: BOOT_POLL_SLEEP, the time to wait between each poll to check if a VM has booted (default is 3s)
• config: New configuration flag: silence-metrics, which will cause metrics not to be printed to the log even if no Librato credentials have been provided
• main: SIGUSR1 is caught and will cause each processor in the pool to print its current status to the logs
• backend: Add --help messages for all backends

Changed

• backend/docker: Container hostnames now begin with travis-docker- instead of travis-go-

Fixed

• step/run-script: Format the timeout duration in the log timeout message as a duration instead of a float

Added

• each Travis CI build will cause three binaries to be uploaded: One for the commit SHA or tag, one for the branch and one for the job number.

Changed

• backend/docker: Improve format of instance ID in the logs for each container

Fixed

• config: Include the build-api-insecure-skip-verify when writing the configuration using --echo-config

Added

• config: New flag: build-api-insecure-skip-verify, which will skip verifying the TLS certificate when requesting the build script

Changed

• config: Hard timeout is now configurable using HARD_TIMEOUT
• backend/docker: Allow for running containers in privileged mode using TRAVIS_WORKER_DOCKER_PRIVILEGED=true
• main: --help will list configuration options
• step/run-script: The instance ID is now printed in the "Using worker" line at the top of the job logs
• backend/docker: Instead of just searching for images tagged with travis:<language>, also search for tags <language>, travis:default and default, in that order
• step/upload-script: Requeue job immediately if a build script has been uploaded, which is a possible indication of a VM being reused

Changed

• backend/jupiter-brain: More options available for image aliases. Now aliases named <osx_image>, osx_image_<osx_image>, osx_image_<osx_image>_<language>, dist_<dist>_<language>, dist_<dist>, group_<group>_<language>, group_<group>, language_<language> and default_<os> will be looked for, in that order.
• logger: The logger always prints key=value formatted logs without colors
• backend/jupiter-brain: Sleep in between requests to check if IP is available

Added

• backend/jupiter-brain

Changed

• backend/docker: CPUs that can be used by containers scales according to number of CPUs available on host
• step/run-script: Print hostname and processor UUID at the top of the job log

Added

• Initial release