Pass in cell to EnsureVSchema and GetOrCreateShard #5930
Conversation
zmagg
force-pushed
the
zmagg_rebuild_srv_schema_no_multicell
branch
2 times, most recently
from
6109866
to
29b8e36
Compare
| @@ -82,10 +83,21 @@ func (ts *Server) EnsureVSchema(ctx context.Context, keyspace string) error { | |||
| } | |||
| } | |||
|
|
|||
| err = ts.RebuildSrvVSchema(ctx, []string{} /* cells */) | |||
There was a problem hiding this comment.
What if we change EnsureVSchema to accept a cells list to rebuild, and we have each tablet rebuild only its own cell? Would that still have caused the problem you had?
There was a problem hiding this comment.
Oh, I like that a lot more. It still does the nice thing for single cell setups, with the benefit of being easier to reason about overall. I'll switch to that approach!
That would've been fine. The problem was that the tablet in the init loop wasn't just rebuilding its own cell it was rebuilding all the other cells.
zmagg
force-pushed
the
zmagg_rebuild_srv_schema_no_multicell
branch
8 times, most recently
from
a70acbf
to
97268cb
Compare
| @@ -54,7 +54,6 @@ func RestoreTablet(t *testing.T, localCluster *cluster.LocalProcessCluster, tabl | |||
| _, err := localCluster.VtctlProcess.ExecuteCommandWithOutput("CreateKeyspace", | |||
| "-keyspace_type=SNAPSHOT", "-base_keyspace="+keyspaceName, | |||
| "-snapshot_time", tm.Format(time.RFC3339), restoreKSName) | |||
| require.Nil(t, err) | |||
There was a problem hiding this comment.
This seems like it should not be removed right?
go/vt/topo/vschema.go
Outdated
| if err != nil { | ||
| log.Errorf("could not rebuild SrvVschema after creating keyspace: %v", err) | ||
| return err | ||
| } | ||
|
|
zmagg
changed the title
| @@ -1620,7 +1620,12 @@ func commandCreateKeyspace(ctx context.Context, wr *wrangler.Wrangler, subFlags | |||
| } | |||
|
|
|||
| if !*allowEmptyVSchema { | |||
| err = wr.TopoServer().EnsureVSchema(ctx, keyspace) | |||
| cells, err := wr.TopoServer().GetKnownCells(ctx) | |||
There was a problem hiding this comment.
I think we could also just pass an empty cells list to mean "all cells", since RebuildSrvVSchema already knows how to expand that by doing this same thing.
zmagg
force-pushed
the
zmagg_rebuild_srv_schema_no_multicell
branch
from
f927513
to
c2deff9
Compare
Signed-off-by: Maggie Zhou <[email protected]>
Signed-off-by: Maggie Zhou <[email protected]>
Signed-off-by: Maggie Zhou <[email protected]>
Remove empty line. Signed-off-by: Maggie Zhou <[email protected]>
Signed-off-by: Maggie Zhou <[email protected]>
Signed-off-by: Maggie Zhou <[email protected]>
zmagg
force-pushed
the
zmagg_rebuild_srv_schema_no_multicell
branch
from
4f5ded2
to
72e75c1
Compare
This test hasn't been flaky. Does it pass for you locally? |
zmagg
force-pushed
the
zmagg_rebuild_srv_schema_no_multicell
branch
from
c0f1f4e
to
aadd900
Compare
throw an error Signed-off-by: Maggie Zhou <[email protected]>
zmagg
force-pushed
the
zmagg_rebuild_srv_schema_no_multicell
branch
from
aadd900
to
f8edec4
Compare
| replicaTabletArgs := commonTabletArg | ||
|
|
There was a problem hiding this comment.
@zmagg and I discussed this. It seems like this was a flaky test. The following change addresses that problem.
…ema_no_multicell Pass in cell to EnsureVSchema and GetOrCreateShard
Description
Modifies
EnsureVSchemaso that we always pass in a cell and it rebuilds SrvVSchema only in the provided cell. This removes its auto-rebuild of all cells automatically when you're creating a new shard or keyspace.Background
During a setup of a new region, we had a misconfigured topology and spinning up a new shard caused us to corrupt all cells in the topology, just as part of booting up a new tablet. For safety reasons, it seems better to decouple spinning up a new shard from rebuilding SrvVSchema on all cells. As talked about in person a few weeks ago--we'll keep the rebuild for the simple case where there's only one cell.
Signed-off-by: Maggie Zhou [email protected]