northd: Fix 100% CPU when northd is paused.

This is a similar patch as [0] but for northd.
ovn-northd needs to recompute next time it resumes from a paused ovn-northd.
However, it should not force an immediate wake-up in every paused
loop, using 100% CPU time.
When ovn-northd receives unixctl command "pause", it should wake-up
once, so it can handle the "pause" in next loop, release the
different locks and force a full recompute next time it becomes active.

[0] Fix 100% CPU when ovn-ic is paused.

Fixes: a8a4463 ("inc-engine: Adjust the force recompute API.")
Signed-off-by: Xavier Simonart <xsimonar@redhat.com>
Acked-by: Mark Michelson <mmichels@redhat.com>
Signed-off-by: Ales Musil <amusil@redhat.com>
(cherry picked from commit 8651a32)

Author: simonartxavier

northd/ovn-northd.c

@@ -1172,7 +1172,7 @@ main(int argc, char *argv[])
             ovsdb_idl_wait(ovnsb_idl_loop.idl);
 
             /* Force a full recompute next time we become active. */
-            inc_proc_northd_force_recompute_immediate();
+            inc_proc_northd_force_recompute();
         }
 
         if (clear_idl_track) {
@@ -1240,6 +1240,7 @@ ovn_northd_pause(struct unixctl_conn *conn, int argc OVS_UNUSED,
     struct northd_state  *state = state_;
     state->paused = true;
 
+    poll_immediate_wake();
     unixctl_command_reply(conn, NULL);
 }
 

tests/ovn-northd.at

@@ -858,7 +858,13 @@ Status: paused
 AS_BOX([Verify that ovn-northd is paused])
 # Now ovn-northd won't respond by adding a datapath, because it's paused.
 check ovn-nbctl ls-add sw0
+cnt1=$(as northd ovn-appctl -t ovn-northd coverage/read-counter poll_zero_timeout)
 check sleep 5
+cnt2=$(as northd ovn-appctl -t ovn-northd coverage/read-counter poll_zero_timeout)
+delta=$((cnt2 - cnt1))
+echo "$cnt1 poll_zero_timeout before sleep(5), and $cnt2 after"
+AT_CHECK([test $((cnt2 - cnt1)) -lt 100])
+
 check_row_count Datapath_Binding 0
 
 # Do not resume both main and backup right after each other