Dumitru reported in #openvswitch that ovn-northd-ddlog can discard the
setting of NB_Global.options:use_logical_dp_groups at startup. I think
that this must be because it seems possible that at startup some of the
relations in the Out_NB_Global rule aren't populated yet, and yet
there is still a row in nb::NB_Global, so that neither rule for
Out_NB_Global matches and therefore ovn-northd-ddlog deletes the row.
This commit changes the structure of how ovn-northd-ddlog generates
Out_NB_Global to ensure that, if there's an input row, it always
generates exactly one output row. This should be more robust than the
previous version regardless of whether it fixes the exact problem
that Dumitru observed (which I did not try to reproduce).
Reported-by: Dumitru Ceara <dce...@redhat.com>
Signed-off-by: Ben Pfaff <b...@ovn.org>
---
northd/ovn_northd.dl | 63 ++++++++++++++++++++++++++------------------
1 file changed, 37 insertions(+), 26 deletions(-)
diff --git a/northd/ovn_northd.dl b/northd/ovn_northd.dl
index 0063021e13f5..d718425b7de3 100644
--- a/northd/ovn_northd.dl
+++ b/northd/ovn_northd.dl
@@ -625,21 +625,25 @@ HvCfgTimestamp(hv_cfg_timestamp) :-
not HvCfgTimestamp0().
/*
- * NB_Global:
- * - set `sb_cfg` to the value of `SB_Global.nb_cfg`.
- * - set `hv_cfg` to the smallest value of `nb_cfg` across all `Chassis`
- * - FIXME: we use ipsec as unique key to make sure that we don't create
multiple `NB_Global`
- * instance. There is a potential race condition if this field is modified
at the same
- * time northd is updating `sb_cfg` or `hv_cfg`.
+ * nb::Out_NB_Global.
+ *
+ * OutNBGlobal0 generates the new record in the common case.
+ * OutNBGlobal1 generates the new record as a copy of nb::NB_Global, if
sb::SB_Global is missing.
+ * nb::Out_NB_Global makes sure we have only a single record in the relation.
+ *
+ * (We don't generate an NB_Global output record if there isn't
+ * one in the input. We don't have enough entropy available to
+ * generate a random _uuid. Doesn't seem like a big deal, because
+ * OVN probably hasn't really been initialized yet.)
*/
-input relation NbCfgTimestamp[integer]
-nb::Out_NB_Global(._uuid = _uuid,
- .sb_cfg = sb_cfg,
- .hv_cfg = hv_cfg,
- .nb_cfg_timestamp = nb_cfg_timestamp,
- .hv_cfg_timestamp = hv_cfg_timestamp,
- .ipsec = ipsec,
- .options = options) :-
+relation OutNBGlobal0[nb::Out_NB_Global]
+OutNBGlobal0[nb::Out_NB_Global{._uuid = _uuid,
+ .sb_cfg = sb_cfg,
+ .hv_cfg = hv_cfg,
+ .nb_cfg_timestamp = nb_cfg_timestamp,
+ .hv_cfg_timestamp = hv_cfg_timestamp,
+ .ipsec = ipsec,
+ .options = options}] :-
NbCfgTimestamp[nb_cfg_timestamp],
HvCfgTimestamp(hv_cfg_timestamp),
nbg in nb::NB_Global(._uuid = _uuid, .ipsec = ipsec),
@@ -654,19 +658,26 @@ nb::Out_NB_Global(._uuid = _uuid,
var options2 = options1.insert_imm("max_tunid", "${max_tunid}"),
var options = options2.insert_imm("northd_internal_version",
ovn_internal_version()).
+relation OutNBGlobal1[nb::Out_NB_Global]
+OutNBGlobal1[x] :- OutNBGlobal0[x].
+OutNBGlobal1[nb::Out_NB_Global{._uuid = nbg._uuid,
+ .sb_cfg = nbg.sb_cfg,
+ .hv_cfg = nbg.hv_cfg,
+ .ipsec = nbg.ipsec,
+ .options = nbg.options,
+ .nb_cfg_timestamp = nbg.nb_cfg_timestamp,
+ .hv_cfg_timestamp = nbg.hv_cfg_timestamp}] :-
+ Unit(),
+ not OutNBGlobal0[_],
+ nbg in nb::NB_Global().
+
+nb::Out_NB_Global[y] :-
+ OutNBGlobal1[x],
+ var y = x.group_by(()).group_first().
-/* SB_Global does not exist yet -- just keep the old value of NB_Global */
-nb::Out_NB_Global(._uuid = nbg._uuid,
- .sb_cfg = nbg.sb_cfg,
- .hv_cfg = nbg.hv_cfg,
- .ipsec = nbg.ipsec,
- .options = nbg.options,
- .nb_cfg_timestamp = nb_cfg_timestamp,
- .hv_cfg_timestamp = hv_cfg_timestamp) :-
- NbCfgTimestamp[nb_cfg_timestamp],
- HvCfgTimestamp(hv_cfg_timestamp),
- nbg in nb::NB_Global(),
- not sb::SB_Global().
+// Tracks the value that should go into NB_Global's 'nb_cfg_timestamp' column.
+// ovn-northd-ddlog.c pushes the current time directly into this relation.
+input relation NbCfgTimestamp[integer]
output relation SbCfg[integer]
SbCfg[sb_cfg] :- nb::Out_NB_Global(.sb_cfg = sb_cfg).
--
2.29.2
_______________________________________________
dev mailing list
d...@openvswitch.org
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
