Closed zimnx closed 1 month ago
Manager flake - https://github.com/scylladb/scylla-operator/issues/2061#issuecomment-2426814613 /retest
/approve
/assign @rzetelskik (I'll be on PTO till Tuesday)
[APPROVALNOTIFIER] This PR is APPROVED
This pull-request has been approved by: tnozicka, zimnx
The full list of commands accepted by this bot can be found here.
The pull request process is described here
/lgtm thanks
Description of your changes:
ScyllaDB, during regular operation, may need to manage millions of open files due to the nature of its workload and architecture. The open file limit (rlimit) for containers is inherited from the CRI and systemd, both of which tend to set conservative limits to avoid misbehavior in other programs when high limits are applied. Simply setting fs.nr_open using ScyllaCluster sysctls API is insufficient to raise these limits for ScyllaDB process.
To automate setting it, Scylla Operator
NodeConfig
container optimization was extended with additional Job discovering the maximum possible limit and setting it on main process of ScyllaDB containers. ScyllaDB Pods await until limit is changed before starting ScyllaDB process. Any forks (sidecar starter or hypervisor) should inherit the limits.Users should increase
fs.nr_open
to at least value recommended by ScyllaDB, because defaults of popular Container Runtimes are ~1024 times lower. Sysctls can currenly be changed viascylladbcluster.spec.sysctls
field. Note that this tuning is applied only on Nodes matching deployed NodeConfig selector.Which issue is resolved by this Pull Request: Resolves #2131