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
{{ message }}
Repository navigation
Commit 18d1ad7
Browse filesBrowse the repository at this point in the historyBrowse files
Handle() returned early when a handler reported an error and silently
dropped the response message the handler had built. The visible case: a
Session Deletion Response was never transmitted when the datapath cleanup
hit "delete: key does not exist" (measured rx=21 vs tx=8 within one
workload), so the CP function (open5gs SMF) saw no response and timed out
its PFCP transaction — on some open5gs versions that aborts the whole SMF
process (open5gs/open5gs#4833).
Changes:
- Handle(): always transmit the handler's response when non-nil; a send
error is logged and only propagated when the handler succeeded.
- mapOperations (cmd/ebpf): delete operations now treat "key absent"
(ebpf.ErrKeyNotExist / ENOENT) as success via isKeyAbsent(), so calling
code does not have to care about this error — the desired end state of a
delete is that the entry is absent. This uniformly covers every delete
call site, including HandlePfcpSessionModificationRequest.
- DeleteUrr() is split: ReadUrr() explicitly returns the URR usage data,
and DeleteUrr() follows the conventional error-only signature.
Verified with a patched build on a session-churn workload: all deletion
requests answered (rx=7/tx=7, RequestAccepted), and no "Error handling
PFCP" warnings in the log.
Fixesedgecomllc#647
Signed-off-by: frank <[email protected]>
0 commit comments