-
Notifications
You must be signed in to change notification settings - Fork 1.3k
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
nonroot install: ubuntu18 failed to start for NOFILE rlimit too low
#7133
Comments
I wonder how we should fix it.
|
|
Hmm. Not sure what to do about it, we rely on the raised limit so much. |
AFAICT, you could change the rlimits for user systemd services in https://bugzilla.redhat.com/show_bug.cgi?id=1364332 However, you cannot go beyond the OS-configured hard limit. |
Hmm, I think we at least want to adjust LimitNOFILE to hard limit value (1048576 in this case), correct? |
It looks like there are two sources of limits: $ grep 'Max open files' /proc/$$/limits
Max open files 1000000 1000000 files
$ systemctl show user@1000.service | grep LimitNOFILE
LimitNOFILE=524288 The first is from |
@avikivity What does it means? |
I'm not sure what it means. We'll have to find out what systemd does. |
I think user@1000.service decreasing NOFILE for user systemd units, at least on Ubuntu 18.
Even hard limit of NOFILE on standard process is 1048576, scylla_prepare reports
|
Ahh, I just posted user@1000.service values on our supported distributions, but I did |
Seems like only new distro has larger value of LimitNOFILE, if it applied as hard limit, we should see CentOS8
Ubuntu16
Ubuntu18
Ubuntu20
Debian9
Debian10:
|
I did one more mistake: I mistakenly had tried reproduce the issue on Ubuntu 16, not 18. I going to try on Ubuntu 18, and also other distros. |
Confirmed able to reproduce the issue on Ubuntu 18, too.
|
We will just have to live with it. Let's make setup warn that there will be limits on the total amount of data. |
This was wrong, CentOS8 has more larger value unlike Ubuntu 18:
I tried to print out NOFILE value in scylla-server service in CentOS8, looks like it comming from user@1000.service, not from scylla-server.service:
On Ubuntu 16 it was:
So I think LimitNOFILE in scylla-server.service is ignored in user mode on all distributions, probably because our setting value I think we have to use developer mode for these distributions. |
…n NOFILE is too low On Ubuntu 16/18 and Debian 9, LimitNOFILE is set to 4096 and not able to override from user unit. To run scylla-server in such environment, we need to turn on developer mode and show warnings. Fixes scylladb#7133
I can still see this issue with scylla-unified-package-4.3.rc2.0.20201123.bc922a743.tar.gz test job: https://jenkins.scylladb.com/job/scylla-4.3/job/artifacts-offline-install/job/artifacts-ubuntu1804-test-nonroot/2/console
|
The fail isn't caused by scylla issue, the problem can be solved by assigning nic name in scylla_setup cmdline. I also sent a patch to SCT:
Let's close this issue. |
Fix present on all active branches, not backporting. |
Installation details
Scylla version (or git commit hash): unified-package-0.20200824.9636a3399.tar.gz
Cluster size: 1
OS (RHEL/CentOS/Ubuntu/AWS AMI): Ubuntu18.
Description
After offline installation on Ubuntu18.04, scylla-server fails to start.
root install works well. and nonroot install of centos8 & Debian 10 works well. This problem only exists with Ubuntu18.04.
I saw scylla-server.service has big limit (
LimitNOFILE=800000
)/CC @roydahan @syuu1228
The text was updated successfully, but these errors were encountered: