'lflow_map' is never expanded because northd always uses fast insertion. This leads to the case where we have a hash map with only 128 initial buckets and every ovn_lflow_find() ends up iterating over n_lflows / 128 entries. It's thousands of logical flows or even more. For example, it takes forever for ovn-northd to start up with the Northbound Db from the 120 node density-heavy test from ovn-heater, because every lookup is slower than previous one. I aborted the process after 15 minutes of waiting, because there was no sign that it will converge. With this change applied the loop completes in only 11 seconds.
Hash map will be pre-allocated to the maximum seen number of logical flows on a second iteration, but this doesn't help for the first iteration when northd first time connects to a big Northbound database, which is a common case during failover or cluster upgrade. And there is an even trickier case where big NbDB transaction that explodes the number of logical flows received on not the first run. We can't expand the hash map in case of parallel build, so this should be fixed separately. CC: Anton Ivanov <anton.iva...@cambridgegreys.com> Fixes: 74daa0607c7f ("ovn-northd: Introduce parallel lflow build") Signed-off-by: Ilya Maximets <i.maxim...@ovn.org> --- northd/ovn-northd.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/northd/ovn-northd.c b/northd/ovn-northd.c index 3d8e21a4f..40cf957c0 100644 --- a/northd/ovn-northd.c +++ b/northd/ovn-northd.c @@ -4387,7 +4387,11 @@ do_ovn_lflow_add(struct hmap *lflow_map, struct ovn_datapath *od, nullable_xstrdup(ctrl_meter), ovn_lflow_hint(stage_hint), where); hmapx_add(&lflow->od_group, od); - hmap_insert_fast(lflow_map, &lflow->hmap_node, hash); + if (!use_parallel_build) { + hmap_insert(lflow_map, &lflow->hmap_node, hash); + } else { + hmap_insert_fast(lflow_map, &lflow->hmap_node, hash); + } } /* Adds a row with the specified contents to the Logical_Flow table. */ -- 2.31.1 _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev