Is your feature request related to a problem? Please describe.
In settingsworker.go we store on disk in the system tenant a cache of settings. This allows us to have some access to all settings before the KV layer becomes available (e.g. during node restarts).
When the settings watcher finds a rangefeed update, we write it
|
return errors.Wrap(storeCachedSettingsKVs(ctx, eng, settingsKVs), "while storing settings kvs") |
The issue is, when we write it, we don't clear out the values which are now unset. See
|
// storeCachedSettingsKVs stores or caches node's settings locally. |
|
// This helps in restoring the node restart with the at least the same settings with which it died. |
|
func storeCachedSettingsKVs(ctx context.Context, eng storage.Engine, kvs []roachpb.KeyValue) error { |
|
batch := eng.NewBatch() |
|
defer batch.Close() |
|
for _, kv := range kvs { |
|
kv.Value.Timestamp = hlc.Timestamp{} // nb: Timestamp is not part of checksum |
|
if err := storage.MVCCPut( |
|
ctx, batch, nil, keys.StoreCachedSettingsKey(kv.Key), hlc.Timestamp{}, kv.Value, nil, |
|
); err != nil { |
|
return err |
|
} |
|
} |
|
return batch.Commit(false /* sync */) |
|
} |
Describe the solution you'd like
We should clear the keys which do exist and are not in the set we want to write.
Additional context
This can be very annoying as we might retain values which are no longer valid. See #68335.
Jira issue: CRDB-10124
Epic: CRDB-6671
Is your feature request related to a problem? Please describe.
In
settingsworker.gowe store on disk in the system tenant a cache of settings. This allows us to have some access to all settings before the KV layer becomes available (e.g. during node restarts).When the settings watcher finds a rangefeed update, we write it
cockroach/pkg/server/settingsworker.go
Line 60 in 9c8aa87
The issue is, when we write it, we don't clear out the values which are now unset. See
cockroach/pkg/server/settings_cache.go
Lines 24 to 38 in 9c8aa87
Describe the solution you'd like
We should clear the keys which do exist and are not in the set we want to write.
Additional context
This can be very annoying as we might retain values which are no longer valid. See #68335.
Jira issue: CRDB-10124
Epic: CRDB-6671