Ruby: Tests frozen and cancelled after 10 minute log silence #. In some cases, the use of the timecop gem can result in seemingly sporadic “freezing” due to issues with ordering calls of Timecop.return, Timecop.freeze, and Timecop.travel. On OS X, verify that homebrew is installed and run: brew install cmake m4 automake peg libtool autoconf. The version of Clang that ships with the OS X toolchain does not support OpenMP, which is a requirement for RedisGraph.
The resulting image is around 7.55MB which should be small enough for our use cases. List of steps how to publish to GitHub Container Registry with GitHub Actions. In order to publish a container image on GitHub Container Registry using GitHub Actions, we have to do the following steps. With docker data volumes it's very easy to expose xorg's unix domain socket inside the container. For example, with a Dockerfile like this: FROM debian RUN apt-get update RUN apt-get install -qqy x11-apps ENV DISPLAY:0 CMD xeyes. Our pre-compiled binary packages and Docker images include a self-contained Python 3 environment, a recent version of OpenSSL that support ALPN and HTTP/2, and other dependencies that would otherwise be cumbersome to compile and install. Dependencies in the binary packages are frozen on release, and can’t be updated in situ.
My tests broke but were working yesterday #
A very common cause when a test is suddenly breaking without any major codechanges involved is a change in upstream dependencies.
This can be an Ubuntu package or any of your project’s language dependencies,like RubyGems, NPM packages, Pip, Composer, etc.
To find out if this is the case:
Restart a build that used to be green, the last known working one, for instance.If that build suddenly fails too, there’s a good chance, that a dependency wasupdated and is causing the breakage.
Check the list of dependencies in the build log, usually outputincluding versions and see, if there’s anything that’s changed.
Sometimes, this can also be caused by an indirect dependency that was updated.
After figuring out which dependency was updated, lock it to the last knownversion.
Additionally, we update our build environment regularly, which brings in newerversions of languages and the running services.
My build script is killed without any error #
Sometimes, you’ll see a build script causing an error and the message inthe log will be something like Killed.
This is usually caused by the script or one of the programs it runs exhaustingthe memory available in the build sandbox, which is currently 3GB. Plus, thereare two cores available, bursted.
Depending on the tool in use, this can be caused by a few things:
- Ruby test suite consuming too much memory
- Tests running in parallel using too many processes or threads (e.g. using the
parallel_testgem) - g++ needing too much memory to compile files, for instance with a lot oftemplates included.
Parallel processes #
For parallel processes running at the same time, try to reduce the number. Morethan two to four processes should be fine, beyond that, resources are likely tobe exhausted.
Ruby processes #
With Ruby processes, check the memory consumption on your local machine, it’slikely to show similar causes. It can be caused by memory leaks or by customsettings for the garbage collector, for instance to delay a sweep for as long aspossible. Dialing these numbers down should help.
My build fails unexpectedly #
One possible cause for builds failing unexpectedly can be calling set -e (also known as set errexit), either directly in your .travis.yml, or sourceing a script which does. This causes any error causing a non-zero return status in your script to stop and fail the build immediately.
Note that using set -e in external scripts does not cause this problem, as the errexit is effective only in the external script.
See also Complex Build Steps.
Another reason could be that the repo setting Clone or import is set to OFF. In this case, no information from the repository is shared and it is possible some builds using private dependencies between repos can break.If you want to avoid the situation when all of your repositories stop sharing dependencies, please go to the repository settings and explicitly set Clone or Import to ON. In this case, your builds keep running as usual.
Segmentation faults from the language interpreter (Ruby, Python, PHP, Node.js, etc.) #
Docker Brew Start Daemon
If your build is failing due to unexpected segmentation faults in the language interpreter, this may be caused by corrupt or invalid caches of your extension codes (gems, modules, etc). This can happen with any interpreted language, such as Ruby, Python, PHP, Node.js, etc.
Fix the problem by
- clearing the cache or
- removing the cache key from your .travis.yml (you can add it back in a subsequent commit).
Ruby: RSpec returns 0 even though the build failed #
In some scenarios, when running rake rspec or even rspec directly, the commandreturns 0 even though the build failed. This is commonly due to some RubyGemoverwriting the at_exit handler of another RubyGem, in this case RSpec’s.
The workaround is to install this at_exit handler in your code, as pointed outin this article.
If your project is using the Code Climate integration orSimplecov, this issue can also come up with the 0.8 branch of Simplecov. The fixis to downgrade to the last 0.7 release until the issue is fixed.
Capybara: I’m getting errors about elements not being found #
In scenarios that involve JavaScript, you can occasionally see errors thatindicate that an element is missing, a button, a link, or some other resourcethat is updated or created by asynchronous JavaScript.
This can indicate that the timeouts used for Selenium or one of its drivers areset too low.
Capybara has a timeout setting which you can increase to a minimum of 15seconds:
Poltergeist has its own setting for timeouts:
If you’re still seeing timeouts after increasing it initially, set it tosomething much higher for one test run. Should the error still persist, there’spossibly a deeper issue on the page, for instance compiling the assets.
Ruby: Installing the debugger_ruby-core-source library fails #
This Ruby library unfortunately has a history of breaking with even patchlevelreleases of Ruby. It’s commonly a dependency of libraries like linecache orother Ruby debugging libraries.
We recommend moving these libraries to a separate group in your Gemfile and thento install RubyGems on Travis CI without this group. As these libraries are onlyuseful for local development, you’ll even gain a speedup during the installationprocess of your build.
Ruby: Tests frozen and cancelled after 10 minute log silence #
In some cases, the use of the timecop gem can result in seemingly sporadic“freezing” due to issues with ordering calls of Timecop.return,Timecop.freeze, and Timecop.travel. For example, if using RSpec, be sure tohave a Timecop.return configured to run after all examples:
Mac: macOS Mavericks (10.9) Code Signing Errors #
With Mavericks quite a lot has changed in terms of code signing and the keychain application.
Signs of issues can be error messages stating that an identity can’t be found and that “Userinteraction is not allowed.”
The keychain must be marked as the default keychain, must be unlocked explicitly and the build needs to make sure that the keychain isn’t locked before the critical point in the build is reached. The following set of commands takes careof this:
Mac: macOS Sierra (10.12) Code Signing Errors #
With the introduction of macOS Sierra (10.12) on our infrastructure, we’ve seen build jobs that were hanging at the codesigning step of the build process. Here’s some information on how to recognize this issue and fix it.
Your build is running on macOS Sierra (10.12), if the osx_image in your .travis.yml file is xcode8.3 or higher. See the macOS Build Environment documentation to know which macOS version is associated with each image.
The following lines in your build log possibly indicate an occurrence of this issue:
Example: Signing
Example: Embed Pods Frameworks
To fix this issue, you will need to add the following command after you have imported your certificate:
Where:
keychainPassis the password of your keychainkeychainNameis the name of your keychain
Here’s an example of where to put the command in context:
IMPORTANT: It’s mandatory to create a keychain with a password for the command security set-key-partition-list to work.
Fastlane #
If you are using Fastlane to sign your app (e.g. with Fastlane Match), you will need to do something similar to the following in your Fastfile:
If you are using import_certificate directly to import your certificates, it’s mandatory to pass your keychain’s password as a parameter e.g.
You can also have more details in this GitHub issue starting at this comment.
Mac: Errors running CocoaPods #
CocoaPods usage can fail for a few reasons currently.
Newer version of CocoaPods required #
Most Pods now require CocoaPods 0.32.1, but we still have 0.21 preinstalled. Ifyou’re seeing this error, add this to your .travis.yml:
CocoaPods can’t be found #
CocoaPods isn’t currently installed on all available Rubies, which unfortunatelymeans it will fail when using the default Ruby, which is 2.0.0.
To work around this issue, you can either install CocoaPods manually as shownabove, or you can switch to Ruby 1.9.3 in your .travis.yml, which should workwithout any issues:
CocoaPods fails with a segmentation fault #
On Ruby 2.0.0, CocoaPods has been seen crashing with a segmentation fault.
You can work around the issue by using Ruby 1.9.3, which hasn’t shown theseissues. Add this to your .travis.yml:
System: Required language pack isn’t installed #
The Travis CI build environments currently have only the en_US language packinstalled. If you get an error similar to : “Error: unsupported localesetting”, then you may need to install another language pack during your testrun.
This can be done with the follow addition to your .travis.yml:
The above addition will reinstall the en_US language pack as well as the de_DElanguage pack.
If you are running on the container-based infrastructure and don’t have accessto the sudo command, install locales using the APT addon:
Linux: apt fails to install package with 404 error #
This is often caused by old package database and can be fixed by adding the following to .travis.yml:
Windows: common build problems and known issues #
For a list of common build problems on Windows, known issues and workarounds, please visit the [Travis CI community forum].(https://travis-ci.community/t/current-known-issues-please-read-this-before-posting-a-new-topic/264).The Travis CI community forum provides better visibility on the issues customers are running into and how to solve them.
Travis CI does not preserve the state between builds #
Travis CI uses virtual machine snapshotting to make sure no state is preserved betweenbuilds. If you modify the CI environment by writing something to a data store, creatingfiles or installing a package via apt, it does not affect subsequent builds.
SSH is not working as expected #
Travis CI runs all commands over SSH in isolated virtual machines. Commands thatmodify SSH session states are “sticky” and persist throughout the build. For example,if you cd into a directory, all subsequent commands are run from that directory.
Git submodules are not updated correctly #
Travis CI automatically initializes and updates submodules when there’s a .gitmodules file in the root of the repository.
To turn this off, set:
If your project requires specific options for your Git submodules, which Travis CIdoes not support out of the box, turn off the automatic integration and use thebefore_install hook to initializes and update them.
For example, to update nested submodules:
Git cannot clone my Submodules #
If your project uses Git submodules, make sure you use public Git URLs. For example, on GitHub, instead of
use
Otherwise, Travis CI builders won’t be able to clone your project because they don’t have your private SSH key.
My builds are timing out #
Builds can unfortunately time out, either during installation of dependencies or during the build itself, for instance because of a command that’s taking a longer amount of time to run while not producing any output.
Our builds have a global timeout and a timeout that’s based on the output. If no output is received from a build for 10 minutes, it’s assumed to have stalled for unknown reasons and is subsequently killed.
At other times, installation of dependencies can timeout. Bundler and RubyGems are a relevant example. Network connectivity between our servers can sometimes affect connectivity to APT, Maven or other repositories.
There are few ways to work around that.
Timeouts installing dependencies #
If you are getting network timeouts when trying to download dependencies, eitheruse the built in retry feature of your dependency manager or wrap your installcommands in the travis_retry function.
Bundler #
Bundler retries three times by default, but if you need to increase that number,use the following syntax in your .travis.yml
travis_retry #
For commands which do not have a built-in retry feature, use the travis_retryfunction to retry it up to three times, if the return code is non-zero:
Most of our internal build commands are wrapped with travis_retry to reduce theimpact of network timeouts.
Note that travis_retry does not work in the deploy step of the build, although itdoes work in the other steps.
Build times out because no output was received #
When a long running command or compile step regularly takes longer than 10 minutes without producing any output, you can adjust your build configuration to take that into consideration.
The shell environment in our build system provides a function that helps to work around that, at least for longer than 10 minutes.
If you have a command that doesn’t produce output for more than 10 minutes, you can prefix it with travis_wait, a function that’s exported by our build environment. For example:
spawns a process running mvn install.travis_wait then writes a short line to the build log every minute for 20 minutes, extending the amount of time your command has to finish.
If you expect the command to take more than 20 minutes, prefix the command with travis_wait n where n is the number of minutes by which the waiting time is extended.
Continuing the example above to extend the waiting time to 30 minutes:
We recommend to carefully use travis_wait, as overusing it can extend your build time when there could be a deeper underlying issue. When in doubt, email us first to see if something could be improved about this particular command first.
Limitations of travis_wait#
travis_wait works by starting a process, sending it to the background, and watching the backgroundprocess.If the command you pass to travis_wait does not persist, then travis_wait does not extend the timeout.
Running builds in debug mode #
In private repositories and those public repositories for which the feature is enabled,it is possible to run builds and jobs in debug mode.Using this feature, you can interact with the live VM where your builds run.
For more information, please consult the debug VM documentation.
Log length exceeded #
The log for each build is limited to approximately 4 MB. When it reaches that length, the build is terminated and you’ll see the following message at the end of your build log:
FTP/SMTP/other protocol do not work #
Some protocols such as FTP and SMTP are not directly supported due to theinfrastructure requirements in place for security and fair usage. Usingalternate stateless
protocols such as HTTPS is best, but tunneling isalso known to work, such as by using SFTP in the specific case of FTP, or a VPNconnection for a wide variety of protocols, e.g.:
I pushed a commit and can’t find its corresponding build #
The build request events that Travis CI receives are listed in your repository’s Requests page. You can find it under the More Options dropdown menu, choosing Requests.
Whenever your build has been processed, you’ll see the message: “Build created successfully”.
If a build hasn’t been triggered for your commit, these are the possible build request messages:
- “Could not authorize build request”, usually means that the account’s subscription expired or that it ran out of trial builds.
- “Build skipped via commit message”, this commit contains the skip command.
- “GitHub payload is missing a merge commit”, please confirm your pull request is open and mergeable. You may also have unresolved conflicts in a particular branch.
- “Branch excluded per configuration” or “Branch not included per configuration”, please make sure your branch is not explicitly excluded or not included in your
.travis.ymlfile. - “Build type disabled via repository settings”, please make sure your Push or Pull Request builds are still active.
- “Build config did not create any jobs.”, please make sure the conditions in your
.travis.ymlfile are able to created a job.
Please note that Travis CI does not receive a Webhook event when more than three commits are tagged. So if you do git push --tags, and more than three tags that are present locally, are not known on GitHub, Travis will not be told about any of those events, and the tagged commits will not be built.
I’m running out of disk space in my build #
Docker Brew Image
Approximate available disk space is listed in the build environment overview.
The best way to find out what is available on your specific image is to run df -h as part of your build script.If you need a bit more space in your Ubuntu builds, we recommend using language: minimal, which will route you to a base image with less tools and languages preinstalled. This image has approximately ~24GB of free space.
Uploading artifacts to sonatype #
When publishing via the nexus-staging-maven-plugin to Sonatype OSS Repository, IP addresses used by TravisCI change due to our NAT layer. To get around this, please use a stagingProfileId as explained in this document.
Travis CLI does not recognize my valid Github token #
When using the Travis CLI tool to interact with the Travis CI platform, if you receive an insufficient_oauth_permissions error or similar, please ensure the Github Token supplied via --github-token has repo scope as explained in this document.
Duplicate/Unknown job shows up in my build #
When specifying stages, users often unknowingly add an implicit job to the list of jobs in a stage using YAML that is otherwise syntactically correct.
The above definition, creates a stage called Breakfast and 2 jobs. The first job is an implicit job that inherits all the default values for the programming language specified. In the example above, the default values for C will be used while the second job is the Peanut Butter and Bread, which you have explicitly defined.
Docker Brew
To remove this implicit job, you would edit the above to look like:
This creates only one job, Peanut Butter and Bread under the stage named Breakfast as you have defined. It is important to note that in YAML, the - symbol is used to create a list of items and the earlier example creates a list of 2 items, while you actually wanted 1. You can read more on How to define Build Stages and YAML lists syntax in the official documentation.
Node: Script execution before dependency installation causes build failures #
When adding custom setup instructions to a NodeJS build, add them in the before_script phase and not before dependencies are installed. The before_script phase is the safest place to add custom setup scripts. Symptoms of this problem include previously succeeding builds suddenly failing due to the addition of a new dependency.