-
Bug
-
Resolution: Done
-
Undefined
-
openshift-4.14.z
-
False
-
None
-
False
-
-
-
-
NetObserv - Sprint 258, NetObserv - Sprint 259
As per the discussion on slack thread: https://redhat-internal.slack.com/archives/C01DMAD88JF/p1724051085870459
Discussion:
- pkt drop hook need special handling for rt kernel.
- netobserv intentionally uses BPF_F_NO_PREALLOC for hashmap to be conservative with memory usage the error from rt nodes seems they didn't like that and wanted to memory to be all preallocated.
- team fixed this issue in rhel9.4 via https://bugzilla.redhat.com/show_bug.cgi?id=2166911 but it wasn't backported to rhel9.2 and fix is there OCP 4.16.
- We checked with RHEL team for fix: https://issues.redhat.com/browse/RHEL-55713 but that was closed with won't do as team currently don't have the capacity to do the backport.
- On thread team suggested to open a Jira here for further work and changes.
2024-08-15T04:41:17.799344034Z time="2024-08-15T04:41:17Z" level=fatal msg="can't instantiate NetObserv eBPF Agent" error="loading and assigning BPF objects: field KfreeSkb: program kfree_skb: load program: invalid argument: trace type programs can only use preallocated hash map (1 line(s) omitted)"
nodes are running realtime kernel version 5.14.0-284.73.1.rt14.358.el9_2.x86_64 while the working nodes are nxts1-osma-001/002/003 and those nodes are running different kernel 5.14.0-284.73.1.el9_2.x86_64
- links to
-
RHSA-2024:135231 Network Observability 1.7.0 for OpenShift
- mentioned on