Details
-
Bug
-
Resolution: Unresolved
-
Critical
-
6.5.0
Description
Clients of the menelaus_event infrastructure are notified of all "interesting" events that occur even if they are only interested in a subset. The "interesting" events consist of:
is_interesting_to_watchers({significant_buckets_change, _}) -> true;
|
is_interesting_to_watchers({memcached, _}) -> true;
|
is_interesting_to_watchers({{node, _, memcached}, _}) -> true;
|
is_interesting_to_watchers({{node, _, membership}, _}) -> true;
|
is_interesting_to_watchers({rebalance_status, _}) -> true;
|
is_interesting_to_watchers({recovery_status, _}) -> true;
|
is_interesting_to_watchers({buckets, _}) -> true;
|
is_interesting_to_watchers({nodes_wanted, _}) -> true;
|
is_interesting_to_watchers({server_groups, _}) -> true;
|
is_interesting_to_watchers({ns_node_disco_events, _NodesBefore, _NodesAfter}) -> true;
|
is_interesting_to_watchers({autocompaction, _}) -> true;
|
is_interesting_to_watchers({cluster_compat_version, _}) -> true;
|
is_interesting_to_watchers({developer_preview_enabled, _}) -> true;
|
is_interesting_to_watchers({cluster_name, _}) -> true;
|
is_interesting_to_watchers({memory_quota, _}) -> true;
|
is_interesting_to_watchers({index_settings_change, memoryQuota, _}) -> true;
|
is_interesting_to_watchers({indexes_change, _, _}) -> true;
|
is_interesting_to_watchers({goxdcr_enabled, _}) -> true;
|
is_interesting_to_watchers({{node, _, stop_xdcr}, _}) -> true;
|
is_interesting_to_watchers({{node, _, services}, _}) -> true;
|
is_interesting_to_watchers({{service_map, _}, _}) -> true;
|
is_interesting_to_watchers({user_roles, _}) -> true;
|
is_interesting_to_watchers({client_cert_auth, _}) -> true;
|
is_interesting_to_watchers({audit_uid_change, _}) -> true;
|
One side effect of this is seen for streaming REST Api handlers which wake up on an event notification, do some sort of heavy weighted operation, determine if anything has changed, and if not hibernate again. There have been cases where there are a significant number of processes that all wake up and do this placing a heavy load on the REST Api handling in the system.
Attachments
Issue Links
- blocks
-
MB-46909 Improve Cluster Management Resource Efficiency
- Closed