You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
13:54:44 fatal error: /Library/Developer/CommandLineTools/usr/bin/libtool: can't write to output file (No space left on device)
13:55:20 make[2]: *** [/Users/iojs/build/workspace/node-test-commit-osx/nodes/osx1015/out/Release/libv8_initializers.a] Error 1
im going to say no, /System/Volumes/Data is the map of the user directory IIRC so its likely is a full disk. Catalina changed this reporting into a whole weird mess.
and (probably obviously from the above) obj.target is quite large at over 6Gb
So we need to either get these expanded or find something else to delete. The disks are showing as around 51Gb at the moment. /var/db is quite large at about 8.7Gb but I don't know whether anything can be done about that. ccache for the iojs user is using 4.5Gb
Note these machines cant really be any bigger as the host they are on started hitting issues when the VMs were a bigger size. (This was the roughly the largest size that was stable)
Is that disk size specifically? Any idea what the issues were? I could understand if RAM/CPU allocation was overdone it would lead to issues but disk seems a little odd.
I think it was due to apples autoupdate feature meaning the host needed enough headroom other it crashes - im sure there was an issue on this in the build repo.
Ive just logged into orka-macos11-x64-2 which is the latest failure and there is 15gb of free space available - surely thats enough to build node is it not? or has the binary really ballooned this past year?
Ive just logged into orka-macos11-x64-2 which is the latest failure and there is 15gb of free space available - surely thats enough to build node is it not? or has the binary really ballooned this past year?
Maybe someone with a mac can verify?
The test CI runs make clean at the end of the job so it's hard to catch peak disk usage, but the release jobs do not and the workspaces for the release machines are in that ballpark, with the arm64 almost double presumably because it builds both x64 and arm64:
$ ssh release-orka-macos10.15-x64-1 du -hs /Users/iojs/build/ws/
17G /Users/iojs/build/ws/
$ ssh release-nearform-macos10.15-x64-1 du -hs /Users/iojs/build/ws/
17G /Users/iojs/build/ws/
$ ssh release-nearform-macos11.0-arm64-1 du -hs /Users/iojs/build/ws/
31G /Users/iojs/build/ws/
and the release job doesn't run tests either so that doesn't account for extra space required by the tests such as test-fs-readfile-tostring-fail.
ah ok well that is a problem then. Also once again the private folder on the machine I rebooted a couple weeks back is over 7.6GB again so this needs some consideration. It might be worth putting in a weekly reboot into these machines (for reference my mac private folder is ~60% of that size and its turned off every day) but thats also only a stop gap.
I think we need to have a think about what to do with these machines and what solutions we can put in place.
test-orka-macos11-x64-2 had a file /System/Volumes/Data/System/Library/Caches/com.apple.coresymbolicationd/data which was 39GB!! have removed, rebooted and brought back on line.
Reacted by Michaël Zasso, Antoine du Hamel, Beth Griggs and Tony Gorez
Personal conclusions
I can confirm the hypothesis made by @AshCripps, seems a solid solution in long term, we can restore a lot of space if we delete the cache files on the machine and reboot.
Just some examples:
test-nearform-macos11.0-arm64-1 has 182G /System/Volumes/Data/System/Library/Caches/com.apple.coresymbolicationd/data.
test-orka-macos11-x64-1 has 36G /System/Library/Caches/com.apple.coresymbolicationd/data
test-orka-macos11-x64-2 has 24G /System/Library/Caches/com.apple.coresymbolicationd/data
Additionally We can also retrieve some marginal gigabytes from macos updates, downloads, etc...
I see several options to overcome this challenge:
We might want to setup some monitorization/metrics that we can track in Grafana in order to we aware of machines with disk space problems (as we do already for infra)
We need some kind of automation to delete cache files and reboot the machines once in a while, but keeping in mind that M1 Nearform machines can be a problem (TBC @mhdawson )
Potential roadmap
I have an old iMac machine that I can repurpose for local testing the following ideas:
This issue is stale because it has been open many days with no activity. It will be closed soon unless the stale label is removed or a comment is made.
Activity
richardlau commentedon Apr 6, 2022
🤷

There have been three node-test-commit-osx builds on the same machine, https://ci.nodejs.org/computer/test-nearform-macos10.15-x64-3/builds, since then which all passed:
And
richardlau commentedon Apr 7, 2022
hmm. https://ci.nodejs.org/job/node-test-commit-osx/44429/nodes=osx1015/console also ran out of space on the same machine, test-nearform-macos10.15-x64-3.
Again we have around 13G of free space.
The test-commit job does a
git cleanat the end. So somehow the job is chewing through 13G while running? (!)richardlau commentedon Apr 7, 2022
@AshCripps do you know if it's normal for
/devand/System/Volumes/Data/hometo be 100% used on macs?AshCripps commentedon Apr 7, 2022
im going to say no,
/System/Volumes/Datais the map of the user directory IIRC so its likely is a full disk. Catalina changed this reporting into a whole weird mess.tniessen commentedon Apr 8, 2022
So far, all three CI runs on nodejs/node#42628 failed with
ENOSPCon osx.richardlau commentedon Apr 8, 2022
I'm just more and more confused:
sxa commentedon Apr 8, 2022
FYI the builds really are using about 14GB while in flight:
During build (while running tests):
After build (Workspace cleared up):
Some of the files in
out/Releaseare pretty big - these two account for 5GB alone;and (probably obviously from the above)
obj.targetis quite large at over 6GbSo we need to either get these expanded or find something else to delete. The disks are showing as around 51Gb at the moment.
/var/dbis quite large at about 8.7Gb but I don't know whether anything can be done about that.ccachefor theiojsuser is using 4.5Gbsxa commentedon Apr 8, 2022
Note - the
-1machine has a slightly larger disk than-2and-3:AshCripps commentedon Apr 8, 2022
Note these machines cant really be any bigger as the host they are on started hitting issues when the VMs were a bigger size. (This was the roughly the largest size that was stable)
sxa commentedon Apr 8, 2022
Is that disk size specifically? Any idea what the issues were? I could understand if RAM/CPU allocation was overdone it would lead to issues but disk seems a little odd.
AshCripps commentedon Apr 9, 2022
I think it was due to apples autoupdate feature meaning the host needed enough headroom other it crashes - im sure there was an issue on this in the build repo.
richardlau commentedon Apr 14, 2022
Jenkins has automatically taken test-nearform-macos10.15-x64-3 offline due to low disk space.
We're now also seeing disk space issues on test-orka-macos11-x64-2,
e.g. https://ci.nodejs.org/job/node-test-commit-osx/44542/nodes=osx11-x64/console
richardlau commentedon Apr 14, 2022
For test-orka-macos11-x64-2 I've removed
/Users/iojs/build/workspace/node-test-commit-osx-as-shared-mdawsonwhich was 1.5GB.15 remaining items
aduh95 commentedon Oct 19, 2022
https://ci.nodejs.org/job/node-test-commit-osx/48075/nodes=osx11-x64/testReport/junit/(root)/test/pummel_test_fs_readfile_tostring_fail/
targos commentedon Oct 21, 2022
https://ci.nodejs.org/job/node-test-commit-osx/48110/nodes=osx1015/testReport/junit/(root)/test/pummel_test_fs_largefile/
targos commentedon Oct 22, 2022
I don't know if it's only because of this issue, but at the moment, macOS CI is almost permanently red.
AshCripps commentedon Oct 24, 2022
Ive just logged into
orka-macos11-x64-2which is the latest failure and there is 15gb of free space available - surely thats enough to build node is it not? or has the binary really ballooned this past year?richardlau commentedon Oct 24, 2022
Maybe someone with a mac can verify?
The test CI runs
make cleanat the end of the job so it's hard to catch peak disk usage, but the release jobs do not and the workspaces for the release machines are in that ballpark, with the arm64 almost double presumably because it builds both x64 and arm64:and the release job doesn't run tests either so that doesn't account for extra space required by the tests such as
test-fs-readfile-tostring-fail.targos commentedon Oct 24, 2022
My
nodefolder takes 16G (arm64 mac)AshCripps commentedon Oct 24, 2022
ah ok well that is a problem then. Also once again the private folder on the machine I rebooted a couple weeks back is over 7.6GB again so this needs some consideration. It might be worth putting in a weekly reboot into these machines (for reference my mac private folder is ~60% of that size and its turned off every day) but thats also only a stop gap.
I think we need to have a think about what to do with these machines and what solutions we can put in place.
AshCripps commentedon Oct 25, 2022
test-orka-macos11-x64-2had a file/System/Volumes/Data/System/Library/Caches/com.apple.coresymbolicationd/datawhich was 39GB!! have removed, rebooted and brought back on line.anonrig commentedon Nov 6, 2022
I got the same error on today's build: #3071
UlisesGascon commentedon Nov 19, 2022
I was doing some research about this topic. Here are some useful logs 😄
Note: I could't test all the osx machines (see: #3084, #3083 and #3034)
Machine Logs
test-nearform-macos10.15-x64-1
Relevant logs (some clean up made)
test-nearform-macos11.0-arm64-1
Relevant logs (some clean up made)
test-orka-macos10.14-x64-1
Relevant logs (some clean up made)
test-orka-macos10.14-x64-2
Relevant logs (some clean up made)
test-orka-macos10.14-x64-3
Relevant logs (some clean up made)
test-orka-macos11-x64-1
Relevant logs (some clean up made)
test-orka-macos11-x64-2
Relevant logs (some clean up made)
rebooting Experiment Logs
By just rebooting a machine the cache folder does not purge itself properly
Relevant logs (some clean up made)
Personal conclusions
I can confirm the hypothesis made by @AshCripps, seems a solid solution in long term, we can restore a lot of space if we delete the cache files on the machine and reboot.
Just some examples:
test-nearform-macos11.0-arm64-1has182G /System/Volumes/Data/System/Library/Caches/com.apple.coresymbolicationd/data.test-orka-macos11-x64-1has36G /System/Library/Caches/com.apple.coresymbolicationd/datatest-orka-macos11-x64-2has24G /System/Library/Caches/com.apple.coresymbolicationd/dataAdditionally We can also retrieve some marginal gigabytes from macos updates, downloads, etc...
I see several options to overcome this challenge:
Potential roadmap
I have an old iMac machine that I can repurpose for local testing the following ideas:
github-actions commentedon Sep 19, 2023
This issue is stale because it has been open many days with no activity. It will be closed soon unless the stale label is removed or a comment is made.