I have the follwing error:
Error: ENOSPC: System limit for number of file watchers reached, watch '/home/ runner/work...
I tried all ways to increase the limit (like ulimit -S -n unlimited, sysctl, etc) but seems to not work, neither with sudo
screnshot
My website has a lot of markdown files (~ 80k) used by gatsby to build the final .htmls.
On my machine I need to increase the file limit, of couse, then works. But in the github actions I can't figure out a way to do this.
My github action workflow.yml
name: Build
on: [push, repository_dispatch]
jobs:
update:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout#v1
- name: Increase file limit
run: sudo sysctl -w fs.file-max=65536
- name: Debug
run: ulimit -a
- name: Set Node.js
uses: actions/setup-node#master
with:
node-version: 12.x
- name: Install dependencies
run: npm install
- name: Build
run: npm run build
I think this could be related to this issue: https://github.com/gatsbyjs/gatsby/issues/17321
It sounds like these GitHub/Expo issues might be the problem:
https://github.com/expo/expo-github-action/issues/20
ENOSPC: System limit for number of file watchers reached
https://github.com/expo/expo-cli/issues/277
Handle ENOSPC error (fs.inotify.max_user_watches reached)
Thanks for testing!
I'm afraid this seems to be a GitHub Action
limitation. That docker image is forcing the
fs.inotify.max_user_watches limit to 524288, but apparently GHA is
overwriting this back to 8192. You can see this happen in a fork of
your repo (when we are done, I'll remove the fork ofc, let me know if
you want to have it removed earlier).
Continuing...
Yes, it's related to a limitation of the environment you are running
Expo CLI in. The metro bundler requires a high amount of listeners
apparently. This fails if the host environment is limiting this. So
technically its an environment issue, but I'm not sure if the CLI can
change anything about this.
I find the limit in GitHub Action personally a little low. Like I
tried to outline in an earlier comment on that CLI issue, the
limitation in other CI vendors is actually set to the default max
listeners. Why they did not do this in GH Actions is unclear, that's
what I try to find out. Might be a configurational issue on their
hands, or an intentional limitation.
... And ...
So, there exists a fix, that seemed to work for me when I tried. What
I did was to follow this guys tip: “Increasing the number of
watchers” — #JNYBGR https://link.medium.com/9Zbt3B4pM0
I then did this in my main action.yml With all the specifics
underneath the dev release
steps:
- uses: actions/checkout#v1
- name: Setup kernel for react native, increase watchers
run: echo fs.inotify.max_user_watches=524288 | sudo tee -a /etc/sysctl.conf && sudo sysctl -p
- name: Run dev release fastlane inside docker action
Please let us know if any of this matches your environment/scenario, and if you find a viable workaround.
UPDATE:
The OP tried fs.inotify.max_user_watches=524288 in his .yaml, and now Gatsby is failing with Error: EMFILE: too many open files open '/home/runner/work/virtualizedfy.gatsby, and NodeJS subsequently crashes with an assertion error:
node[3007]: ../src/spawn_sync.cc:460:v8:Maybe<bool> node:SycProcessRunner::TryInitializeAndRunLoop(v8:Local<v8::Value>): Assertion `{uv_loop_init(vu_loop_ == (0)' failed.
ADDITIONAL SUGGESTION:
https://github.com/gatsbyjs/gatsby/issues/12011
Google seems to suggest https://github.com/isaacs/node-graceful-fs as
a drop in replacement for fs, I might also experiment with that to see
if it makes a difference.
EDIT: I can confirm that monkeypatching fs with graceful-fs at the top
of gatsby-node.js as in the snippet below fixes the issue for me.
const realFs = require('fs')
const gracefulFs = require('graceful-fs')
gracefulFs.gracefulify(realFs)
EDIT2: Actually after upgrading from Node 10 to Node 11 everything
seems to be fine without having to patch fs... So all is well!
Related
I have a gitlab job that does not seem to update the repository before being run. Sometimes it leaves some files in their old states and run the script... Any idea ?
For instance when I have a
packagePython:
stage: package
script:
- .\scripts\PackagePython.ps1
tags:
- myServer
cache:
paths:
- .\python\cache\
only:
changes:
- python/**/*
I finally managed to understand what was happening :
I realised that the gitlab-runner did not use exactly the same path for each run on my server, and my script assumed that it did... So I ended up pointing on a build made on the wrong path.
I guess if you think that it is not updating the repository (like I did) make sure you are not referencing hardcoded path/package in your scripts that could refer to previous versions !
For all the other stackoverflow questions, it seems like people are asking either about a private npm git repository or about a different technology stack. I'm pretty sure I can use a private npm registry with GAE Flexible, but I was wondering if it was possible with the Standard version?
From the GAE standard docs, doesn't seem like it is possible. Anyone else figure out otherwise?
Google marked this feature request as "won't fix, intended behavior" but there is a workaround.
Presumably you have access to the environment variables during the build stage of your CI/CD pipeline. Begin that stage by having your build script overwrite the .npmrc file using the value of the environment variable (note: the value, not the variable name). The .npmrc file (and the token in it) will then be available to the rest of the CI/CD pipeline.
For example:
- name: Install and build
env:
NPM_AUTH_TOKEN: ${{ secrets.PRIVATE_REPO_PACKAGE_READ_TOKEN }}
run: |
# Remove these 'echo' statements after we migrate off of Google App Engine.
# See replies 14 and 18 here: https://issuetracker.google.com/issues/143810864?pli=1
echo "//npm.pkg.github.com/:_authToken=${NPM_AUTH_TOKEN}" > .npmrc
echo "#organizationname:registry=https://npm.pkg.github.com" >> .npmrc
echo "always-auth=true" >> .npmrc
npm install
npm run compile
npm run secrets:get ${{ secrets.YOUR_GCP_PROJECT_ID }}
Hat tip to the anonymous heroes who wrote replies 14 and 18 in the Issure Tracker thread - https://issuetracker.google.com/issues/143810864?pli=1
If you have a .npmrc file checked in with your project's code, you would be wise to put a comment at the top, explaining that it will be overwritten during the CI/CD pipeline. Otherwise, Murphy's Law dictates that you (or a teammate) will check in a change to that .npmrc file and then waste an unbounded amount of time trying to figure out why that change has no effect during deployment.
I am running a Node.js app on Google App Engine, using the following command to deploy my code:
gcloud app deploy --stop-previous-version
My desired behavior is for all instances running previous versions to be terminated, but they always seem to stick around. Is there something I'm missing?
I realize they are not receiving traffic, but I am still paying for them and they cause some background telemetry noise. Is there a better way of running this command?
Example output of the gcloud app instances list:
As you can see I have two different versions running.
We accidentally blew through our free Google App Engine credit in less than 30 days because of an errant flexible instance that wasn't cleared by subsequent deployments. When we pinpointed it as the cause it had scaled up to four simultaneous instances that were basically idling away.
tl;dr: Use the --version flag when deploying to specify a version name. An existing instance with the same version will be
replaced then next time you deploy.
That led me down the rabbit hole that is --stop-previous-version. Here's what I've found out so far:
--stop-previous-version doesn't seem to be supported anymore. It's mentioned under Flags on the gcloud app deploy reference page, but if you look at the top of the page where all the flags are listed, it's nowhere to be found.
I tried deploying with that flag set to see what would happen but it seemingly had no effect. A new version was still created, and I still had to go in and manually delete the old instance.
There's an open Github issue on the gcloud-maven-plugin repo that specifically calls this out as an issue with that plugin but the issue has been seemingly ignored.
At this point our best bet at this point is to add --version=staging or whatever to gcloud deploy app. The reference docs for that flag seem to indicate that that it'll replace an existing instance that shares that "version":
--version=VERSION, -v VERSION
The version of the app that will be created or replaced by this deployment. If you do not specify a version, one will be generated for you.
(emphasis mine)
Additionally, Google's own reference documentation on app.yaml (the link's for the Python docs but it's still relevant) specifically calls out the --version flag as the "preferred" way to specify a version when deploying:
The recommended approach is to remove the version element from your app.yaml file and instead, use a command-line flag to specify your version ID
As far as I can tell, for Standard Environment with automatic scaling at least, it is normal for old versions to remain "serving", though they should hopefully have zero instances (even if your scaling configuration specifies a nonzero minimum). At least that's what I've seen. I think (I hope) that those old "serving" instances won't result in any charges, since billing is per instance.
I know most of the above answers are for Flexible Environment, but I thought I'd include this here for people who are wondering.
(And it would be great if someone from Google could confirm.)
I had same problem as OP. Using the flex environment (some of this also applies to standard environment) with Docker (runtime: custom in app.yaml) I've finally solved this! I tried a lot of things and I'm not sure which one fixed it (or whether it was a combination) so I'll list the things I did here, the most likely solutions being listed first.
SOLUTION 1) Ensure that cloud storage deletes old versions
What does cloud storage have to do with anything? (I hear you ask)
Well there's a little tooltip (Google Cloud Platform Web UI (GCP) > App Engine > Versions > Size) that when you hover over it says:
(Google App Engine) Flexible environment code is stored and billed from Google Cloud Storage ... yada yada yada
So based on this info and this answer I visited GCP > Cloud Storage > Browser and found my storage bucket AND a load of other storage buckets I didn't know existed. It turns out that some of the buckets store cached cloud functions code, some store cached docker images and some store other cached code/stuff (you can tell which is which by browsing the buckets).
So I added a deletion policy to all the buckets (except the cloud functions bucket) as follows:
Go to GCP > Cloud Storage > Browser and click the link (for the relevant bucket) in the Lifecycle Rules column > Click ADD A RULE > THEN:
For SELECT ACTION choose "Delete Object" and click continue
For SELECT OBJECT choose "Number of newer versions" and enter 1 in the input
Click CREATE
This will return you to the table view and you should now see the rule in the lifecycle rules column.
REPEAT this process for all relevant buckets (the relevant buckets were described earlier).
THEN delete the contents of the relevant buckets. WARNING: Some buckets warn you NOT to delete the bucket itself, only the contents!
Now re-deploy and your latest version should now get deployed and hopefully you will never have this problem again!
SOLUTION 2) Use deploy flags
I added these flags
gcloud app deploy --quiet --promote --stop-previous-version
This probably doesn't help since these flags seem to be the default but worth adding just in case.
Note that for the standard environment only (I heard on the grapevine) you can also use the --no-cache flag which might help but with flex, this flag caused the deployment to fail (when I tried).
SOLUTION 3)
This probably does not help at all, but I added:
COPY app.yaml .
to the Dockerfile
TIP 1)
This is probably more of a helpful / useful debug approach than a fix.
Visit GCP > App Engine > Versions
This shows all versions of your app (1 per deployment) and it also shows which version each instance is running (instances are configured in app.yaml).
Make sure all instances are running the latest version. This should happen by default. Probably worth deleting old versions.
You can determine your version from the gcloud app deploy logs (at the start of the logs) but it seems that the versions are listed by order of deployment anyway (most recent at top).
TIP 2)
Visit GCP > App Engine > Instances
SSH into an instance. This is just a matter of clicking a few buttons (see screenshot below). Once you have SSH'd in run:
docker exec -it gaeapp /bin/bash
Which will get you into the docker container running your code. Now you can browse around to make sure it has your latest code.
Well I think my answer is long enough now. If this helps, don't thank me, J-ES-US is the one you should thank ;) I belong to Him ^^
Google may have updated their documentation cited in #IAmKale's answer
Note that if the version is running on an instance of an auto-scaled service, using --stop-previous-version will not work and the previous version will continue to run because auto-scaled service instances are always running.
Seems like that flag only works with manually scaled services.
This is a supplementary and optional answer in addition to my other main answer.
I am now, in addition to my other answer, auto incrementing version manually on deploy using a script.
My script contents are below.
Basically, the script auto increments version every time you deploy. I am using node.js so the script uses npm version to bump the version but this line could easily be tweaked to whatever language you use.
The script requires a clean git working directory for deployment.
The script assumes that when the version is bumped, this will result in file changes (e.g. changes to package.json version) that need pushing.
The script essentially tries to find your SSH key and if it finds it then it starts an SSH agent and uses your SSH key to git commit and git push the file changes. Else it just does a git commit without a push.
It then does a deploy using the --version flag ... --version="${deployVer}"
Thought this might help someone, especially since the top answer talks a lot about using the --version flag on a deploy.
#!/usr/bin/env bash
projectName="vehicle-damage-inspector-app-engine"
# Find SSH key
sshFile1=~/.ssh/id_ed25519
sshFile2=~/Desktop/.ssh/id_ed25519
sshFile3=~/.ssh/id_rsa
sshFile4=~/Desktop/.ssh/id_rsa
if [ -f "${sshFile1}" ]; then
sshFile="${sshFile1}"
elif [ -f "${sshFile2}" ]; then
sshFile="${sshFile2}"
elif [ -f "${sshFile3}" ]; then
sshFile="${sshFile3}"
elif [ -f "${sshFile4}" ]; then
sshFile="${sshFile4}"
fi
# If SSH key found then fire up SSH agent
if [ -n "${sshFile}" ]; then
pub=$(cat "${sshFile}.pub")
for i in ${pub}; do email="${i}"; done
name="Auto Deploy ${projectName}"
git config --global user.email "${email}"
git config --global user.name "${name}"
echo "Git SSH key = ${sshFile}"
echo "Git email = ${email}"
echo "Git name = ${name}"
eval "$(ssh-agent -s)"
ssh-add "${sshFile}" &>/dev/null
sshKeyAdded=true
fi
# Bump version and git commit (and git push if SSH key added) and deploy
if [ -z "$(git status --porcelain)" ]; then
echo "Working directory clean"
echo "Bumping patch version"
ver=$(npm version patch --no-git-tag-version)
git add -A
git commit -m "${projectName} version ${ver}"
if [ -n "${sshKeyAdded}" ]; then
echo ">>>>> Bumped patch version to ${ver} with git commit and git push"
git push
else
echo ">>>>> Bumped patch version to ${ver} with git commit only, please git push manually"
fi
deployVer="${ver//"."/"-"}"
gcloud app deploy --quiet --promote --stop-previous-version --version="${deployVer}"
else
echo "Working directory unclean, please commit changes"
fi
For node.js users if you call the script deploy.sh you should add:
"deploy": "sh deploy.sh"
In your package.json scripts and deploy with npm run deploy
I have a simple Dockerfile
FROM haskell:8
WORKDIR "/root"
CMD ["/bin/bash"]
which I run mounting pwd folder to "/root". In my current folder I have a Haskell project that uses stack (funblog). I configured in stack.yml to use "lts-7.20" resolver, which aims to install ghc-8.0.1.
Inside the container, after running "stack update", I ran "stack setup" but I am getting "Too many open files in system" during GHC compilation.
This is my stack.yaml
flags: {}
packages:
- '.'
- location:
git: https://github.com/agrafix/Spock.git
commit: 2c60a48b2c0be0768071cc1b3c7f14590ffcc7d6
subdirs:
- Spock
- Spock-core
- reroute
- location:
git: https://github.com/agrafix/Spock-digestive.git
commit: 4c85647427e21bbaefbf04c4bc315d4bdfabba0e
extra-deps:
- digestive-bootstrap-0.1.0.1
- blaze-bootstrap-0.1.0.1
- digestive-functors-blaze-0.6.0.6
resolver: lts-7.20
One import note: I don't want to use Docker to deploy the app, just to compile it, i.e. as part of my dev process.
Any ideas?
Should I use another image without ghc pre-installed to use with docker? Which one?
update
Yes, I could use the built-in GHC in the container and it is a good idea, but wondered if there is any issue building GHC within Docker.
update 2
For anyone wishing to reproduce (on MAC OSX by the way), you can clone repo https://github.com/carlosayam/funblog and grab this commit 9446bc0e52574cc574a9eb5f2733f69e07b874ef
(I will probably move on using container's GHC)
By default, Docker for macOS limits number of file descriptors to avoid hitting macOS system-wide limits (default limit is 900). To increase the limit, follow these commands:
$ cd ~/Library/Containers/com.docker.docker/Data/database/
$ git reset --hard
HEAD is now at 9410b78 last-start-time changed at 1480947038
$ cat com.docker.driver.amd64-linux/slirp/max-connections
900
$ echo 1200 > com.docker.driver.amd64-linux/slirp/max-connections
$ git add com.docker.driver.amd64-linux/slirp/max-connections
$ git commit -s -m 'Update the maximum number of connections'
[master 227a248] Update the maximum number of connections
1 file changed, 1 insertion(+), 1 deletion(-)
Then check the notice messages by:
$ syslog -k Sender Docker
<Notice>: updating connection limit to 1200
To check how many files you got open, run: sysctl kern.num_files.
To check what's your current limit, run: sysctl kern.maxfiles.
To increase it system-wide, run: sysctl -w kern.maxfiles=20480.
Source: Containers become unresponsive due to "too many connections".
See also: Docker: How to increase number of open files limit.
On Linux, you can also try to run Docker with --ulimit, e.g.
docker run --ulimit nofile=5000:5000 <image-tag>
Source: Docker error: too many open files
I want to set up a build pipeline in Concourse for my web application. The application is built using Node.
The plan is to do something like this:
,-> build style guide -> dockerize
source code -> npm install -> npm test -|
`-> build website -> dockerize
The problem is, after npm install, a new container is created so the node_modules directory is lost. I want to pass node_modules into the later tasks but because it is "inside" the source code, it doesn't like it and gives me
invalid task configuration:
you may not have more than one input or output when one of them has a path of '.'
Here's my jobs set up
jobs:
- name: test
serial: true
disable_manual_trigger: false
plan:
- get: source-code
trigger: true
- task: npm-install
config:
platform: linux
image_resource:
type: docker-image
source: {repository: node, tag: "6" }
inputs:
- name: source-code
path: .
outputs:
- name: node_modules
run:
path: npm
args: [ install ]
- task: npm-test
config:
platform: linux
image_resource:
type: docker-image
source: {repository: node, tag: "6" }
inputs:
- name: source-code
path: .
- name: node_modules
run:
path: npm
args: [ test ]
Update 2016-06-14
Inputs and outputs are just directories. So you put what you want output into an output directory and you can then pass it to another task in the same job. Inputs and Outputs can not overlap, so in order to do it with npm, you'd have to either copy node_modules, or the entire source folder from the input folder to an output folder, then use that in the next task.
This doesn't work between jobs though. Best suggestion I've seen so far is to use a temporary git repository or bucket to push everything up. There has to be a better way of doing this since part of what I'm trying to do is avoid huge amounts of network IO.
There is a resource specifically designed for this use case of npm between jobs. I have been using it for a couple of weeks now:
https://github.com/ymedlop/npm-cache-resource
It basically allow you to cache the first install of npm and just inject it as a folder into the next job of your pipeline. You could quite easily setup your own caching resources from reading the source of that one as well, If you want to cache more than node_modules.
I am actually using this npm-cache-resource in combination with a Nexus proxy to speed up the initial npm install further.
Be aware that some npm packages have native bindings that need to be built with the standardlibs that matches the containers linux versions standard libs so, If you move between different types of containers a lot you may experience some issues with libmusl etc, in that case I recommend either streamlinging to use the same container types through the pipeline or rebuilding the node_modules in question...
There is a similar one for gradle (on which the npm one is based upon)
https://github.com/projectfalcon/gradle-cache-resource
This doesn't work between jobs though.
This is by design. Each step (get, task, put) in a Job is run in an isolated container. Inputs and outputs are only valid inside a single job.
What connects Jobs is Resources. Pushing to git is one way. It'd almost certainly be faster and easier to use a blob store (eg S3) or file store (eg FTP).