-
Notifications
You must be signed in to change notification settings - Fork 181
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
EOS leaks filedescriptors #268
Comments
To force it happen earlier, run the acceptance tests, on my Ubuntu20.04 system I cannot run all UI tests without increasing the open-file limits |
I am experimenting a bit with prometheus open fdsgoroutines |
see also #222 |
Reason: OIDC-Requests originating from reva are keep-alive and not reused/cleaned up => Proxy keeps connection open indefinitely. Fix: cs3org/reva#787 We probably also should harden the proxy against such cases. |
@IljaN is this done ? any module to update in ocis ? |
the fix that was done in ocis-reva is already present on the ocis master branch. let me know if there are other fixes to get through updates |
Fixed as https://github.com/cs3org/reva/tree/9b9f2e5af0e9216d59552f94e13416dad4dcc457/pkg is in the latest ocis-reva release which is in turn in the latest ocis release. |
@jnweiger Mind to re-test? |
retested with
Two patterns are recognizable in this system:
Example snippet from an lsof grep:
|
Connecting, disconnecting clients, syncing, removing syncs, uploading files over the next hour has no effect on the number of xrootd fildescriptors.
|
Reproducable. This is a differnet system. it was used for manual testing today:
|
also locally the number of file descriptors is dropping again. I am using this command to list the number of open files per process:
you should replace the final grep with whatever the binary is called. mine is called
all back to normal ... on ocis storage ... testing with owncloud ... |
@butonic We have also prometheus and grafana on ocis.owncloud.works I checked it, everything normal. |
glauth and settings leave grpc connections open, which is intended. |
@micbar ok so only eos leaks? make title more precise, because it is not ocis that leaks. |
@ishank011 @labkode do you see this file descriptor leak as well? AFAICT some xrootd processes are not killed... |
@jnweiger cs3org/reva#1260 this PR tackles all known fd leak known so far. Changes are currently on master so we could give this a try again? :) |
reopen if still an issue. |
Test setup via https://gitea.owncloud.services/jw/hetzner/src/branch/master/make_ocis_test.sh
client connected via https://localhost:9200
Letting this sit for several hours, the logfile suddenly starts scrolling with
Using lsof, it can be seen, that ca 20 processes opened every fildescriptor number from 0 to 1023. There seems to be a per process limit of 1024. OK.
All connections are idle, no client activity during the night. Ocis should close unused filedescriptors.
In this state, the system accepts connectsions, but then hangs:
Incoming connections cause no log messages on the console output.
The text was updated successfully, but these errors were encountered: