Implement: CLUSTER REPLICATE NO ONE #1674
Conversation
skolosov-snap
force-pushed
the
skolosov/replicate-no-one
branch
from
91589d1 to
ff96c0f
Compare
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## unstable #1674 +/- ##
============================================
+ Coverage 71.04% 71.08% +0.04%
============================================
Files 121 123 +2
Lines 65254 65548 +294
============================================
+ Hits 46357 46596 +239
- Misses 18897 18952 +55
|
src/cluster_legacy.c
Outdated
| if (server.primary != NULL) { | ||
| replicationUnsetPrimary(); | ||
| } |
There was a problem hiding this comment.
All other invocation of replicationUnsetPrimary() don't have this wrapped under the condition. Is it unnecessary to invoke it if server.primary is NULL ?
There was a problem hiding this comment.
Also unsure why clusterPromoteSelfToPrimary was introduced, seems like it's the same behavior at this point but good to call this abstraction if there will be additional steps introduced in the future for cluster mode.
There was a problem hiding this comment.
Also unsure why clusterPromoteSelfToPrimary was introduced, seems like it's the same behavior at this point but good to call this abstraction if there will be additional steps introduced in the future for cluster mode.
Ok.
There was a problem hiding this comment.
All other invocation of replicationUnsetPrimary() don't have this wrapped under the condition. Is it unnecessary to invoke it if server.primary is NULL ?
Probably not needed. It looks like left over from KeyDB's implementation where primary needed to be passed to replicationUnsetMaster
skolosov-snap
force-pushed
the
skolosov/replicate-no-one
branch
2 times, most recently
from
fde1ab6 to
4abbae8
Compare
Updated. |
skolosov-snap
force-pushed
the
skolosov/replicate-no-one
branch
from
4abbae8 to
85238e6
Compare
|
The CI job "DCO" is failing. You need to use Why we need it? See here: https://github.com/valkey-io/valkey/blob/unstable/CONTRIBUTING.md#developer-certificate-of-origin thanks! |
skolosov-snap
force-pushed
the
skolosov/replicate-no-one
branch
from
85238e6 to
3789227
Compare
Signed-off-by: Sergey Kolosov <[email protected]>
skolosov-snap
force-pushed
the
skolosov/replicate-no-one
branch
from
3789227 to
e4e8b24
Compare
Done |
| /* Lookup the specified node in our table. */ | ||
| if (c->argc == 4) { | ||
| if (0 != strcasecmp(c->argv[2]->ptr, "NO") || 0 != strcasecmp(c->argv[3]->ptr, "ONE")) { |
There was a problem hiding this comment.
we don't do this 0 != xxx trick, let's keep the same style for the code
| /* Reset manual failover state. */ | ||
| resetManualFailover(); |
There was a problem hiding this comment.
it become a primary, do we need this reset? it has no way to enter the failover check logic. or at least we remove the comment since this line is super easy.
| /* Reset manual failover state. */ | |
| resetManualFailover(); | |
| resetManualFailover(); |
| int empty_db_flags = server.repl_replica_lazy_flush ? EMPTYDB_ASYNC : EMPTYDB_NO_FLAGS; | ||
| emptyData(-1, empty_db_flags, NULL); |
There was a problem hiding this comment.
| int empty_db_flags = server.repl_replica_lazy_flush ? EMPTYDB_ASYNC : EMPTYDB_NO_FLAGS; | |
| emptyData(-1, empty_db_flags, NULL); | |
| emptyData(-1, server.repl_replica_lazy_flush ? EMPTYDB_ASYNC : EMPTYDB_NO_FLAGS, NULL); |
| addReply(c, shared.ok); | ||
| return 1; | ||
| } | ||
| serverLog(LL_NOTICE, "Stop replication and turning myself into empty primary."); |
There was a problem hiding this comment.
let's also use catClientInfoShortString to get the client info for a better audit.
Currently, ValKey doesn't allow to detach replica attached to primary node. So, if you want to change cluster topology the only way to do it is to reset (
CLUSTER RESETcommand) the node. However, this results into removing node from the cluster what affects clients. All clients will keep sending traffic to this node (with getting inaccurate responses) until they refresh their topology.In this change we implement supporting of new argument for CLUSTER REPLICATE command:
CLUSTER REPLICATE NO ONE. When calling this command the node will be converted from replica to empty primary node but still staying in the cluster. Thus, all traffic coming from the clients to this node can be redirected to correct node.