| Message ID | 20230525133745.314125-1-bnemeth@redhat.com |
|---|---|
| State | Changes Requested |
| Headers | show |
| Series | [ovs-dev,v12] ofproto-dpif-upcall: Don't set statistics to 0 when they jump back | expand |
| Context | Check | Description |
|---|---|---|
| ovsrobot/apply-robot | success | apply and check: success |
| ovsrobot/github-robot-_Build_and_Test | fail | github build: failed |
On 5/25/23 15:37, Balazs Nemeth wrote: > The only way that stats->{n_packets,n_bytes} would decrease is due to an > overflow, or if there are bugs in how statistics are handled. In the > past, there were multiple issues that caused a jump backward. A > workaround was in place to set the statistics to 0 in that case. When > this happened while the revalidator was under heavy load, the workaround > had an unintended side effect where should_revalidate returned false > causing the flow to be removed because the metric it calculated was > based on a bogus value. Since many of those bugs have now been > identified and resolved, there is no need to set the statistics to 0. In > addition, the (unlikely) overflow still needs to be handled > appropriately. If an unexpected jump does happen, just log it as a > warning. > > Signed-off-by: Balazs Nemeth <bnemeth@redhat.com> > --- > ofproto/ofproto-dpif-upcall.c | 29 +++++++++++++++++++++++------ > 1 file changed, 23 insertions(+), 6 deletions(-) > > diff --git a/ofproto/ofproto-dpif-upcall.c b/ofproto/ofproto-dpif-upcall.c > index cd57fdbd9..4c34d695e 100644 > --- a/ofproto/ofproto-dpif-upcall.c > +++ b/ofproto/ofproto-dpif-upcall.c > @@ -2339,6 +2339,26 @@ exit: > return result; > } > > +static void > +log_unexpected_stats_jump(struct udpif_key *ukey, > + const struct dpif_flow_stats *stats) We need to add a thread-safety annotation here to make clang happy: OVS_REQUIRES(ukey->mutex) > +{ > + static struct vlog_rate_limit rll = VLOG_RATE_LIMIT_INIT(1, 5); > + struct ds ds = DS_EMPTY_INITIALIZER; > + struct ofpbuf *actions; > + > + odp_format_ufid(&ukey->ufid, &ds); > + ds_put_cstr(&ds, " "); > + odp_flow_key_format(ukey->key, ukey->key_len, &ds); > + ds_put_cstr(&ds, ", actions:"); > + actions = ovsrcu_get(struct ofpbuf *, &ukey->actions); > + format_odp_actions(&ds, actions->data, actions->size, NULL); > + VLOG_WARN_RL(&rll, "Unexpected jump in packet stats from %"PRIu64 > + " to %"PRIu64" when handling ukey %s", > + stats->n_packets, ukey->stats.n_packets, ds_cstr(&ds)); And stats should be in the opposite order. The jump is from ukey->stats.n_packets to stats->n_packets, not the other way around. Sorry I missed that before. Best regards, Ilya Maximets.
diff --git a/ofproto/ofproto-dpif-upcall.c b/ofproto/ofproto-dpif-upcall.c index cd57fdbd9..4c34d695e 100644 --- a/ofproto/ofproto-dpif-upcall.c +++ b/ofproto/ofproto-dpif-upcall.c @@ -2339,6 +2339,26 @@ exit: return result; } +static void +log_unexpected_stats_jump(struct udpif_key *ukey, + const struct dpif_flow_stats *stats) +{ + static struct vlog_rate_limit rll = VLOG_RATE_LIMIT_INIT(1, 5); + struct ds ds = DS_EMPTY_INITIALIZER; + struct ofpbuf *actions; + + odp_format_ufid(&ukey->ufid, &ds); + ds_put_cstr(&ds, " "); + odp_flow_key_format(ukey->key, ukey->key_len, &ds); + ds_put_cstr(&ds, ", actions:"); + actions = ovsrcu_get(struct ofpbuf *, &ukey->actions); + format_odp_actions(&ds, actions->data, actions->size, NULL); + VLOG_WARN_RL(&rll, "Unexpected jump in packet stats from %"PRIu64 + " to %"PRIu64" when handling ukey %s", + stats->n_packets, ukey->stats.n_packets, ds_cstr(&ds)); + ds_destroy(&ds); +} + /* Verifies that the datapath actions of 'ukey' are still correct, and pushes * 'stats' for it. * @@ -2372,18 +2392,15 @@ revalidate_ukey(struct udpif *udpif, struct udpif_key *ukey, push.used = stats->used; push.tcp_flags = stats->tcp_flags; - push.n_packets = (stats->n_packets > ukey->stats.n_packets - ? stats->n_packets - ukey->stats.n_packets - : 0); - push.n_bytes = (stats->n_bytes > ukey->stats.n_bytes - ? stats->n_bytes - ukey->stats.n_bytes - : 0); + push.n_packets = stats->n_packets - ukey->stats.n_packets; + push.n_bytes = stats->n_bytes - ukey->stats.n_bytes; if (stats->n_packets < ukey->stats.n_packets && ukey->stats.n_packets < UINT64_THREE_QUARTERS) { /* Report cases where the packet counter is lower than the previous * instance, but exclude the potential wrapping of an uint64_t. */ COVERAGE_INC(ukey_invalid_stat_reset); + log_unexpected_stats_jump(ukey, stats); } if (need_revalidate) {
The only way that stats->{n_packets,n_bytes} would decrease is due to an overflow, or if there are bugs in how statistics are handled. In the past, there were multiple issues that caused a jump backward. A workaround was in place to set the statistics to 0 in that case. When this happened while the revalidator was under heavy load, the workaround had an unintended side effect where should_revalidate returned false causing the flow to be removed because the metric it calculated was based on a bogus value. Since many of those bugs have now been identified and resolved, there is no need to set the statistics to 0. In addition, the (unlikely) overflow still needs to be handled appropriately. If an unexpected jump does happen, just log it as a warning. Signed-off-by: Balazs Nemeth <bnemeth@redhat.com> --- ofproto/ofproto-dpif-upcall.c | 29 +++++++++++++++++++++++------ 1 file changed, 23 insertions(+), 6 deletions(-)