shalinmangar commented on a change in pull request #1387: SOLR-14210: Include
replica health in healtcheck handler
URL: https://github.com/apache/lucene-solr/pull/1387#discussion_r402303966
##########
File path:
solr/core/src/java/org/apache/solr/handler/admin/HealthCheckHandler.java
##########
@@ -88,15 +96,46 @@ public void handleRequestBody(SolrQueryRequest req,
SolrQueryResponse rsp) throw
return;
}
- // Set status to true if this node is in live_nodes
- if
(clusterState.getLiveNodes().contains(cores.getZkController().getNodeName())) {
- rsp.add(STATUS, OK);
- } else {
+ // Fail if not in live_nodes
+ if
(!clusterState.getLiveNodes().contains(cores.getZkController().getNodeName())) {
rsp.add(STATUS, FAILURE);
rsp.setException(new
SolrException(SolrException.ErrorCode.SERVICE_UNAVAILABLE, "Host Unavailable:
Not in live nodes as per zk"));
+ return;
}
- rsp.setHttpCaching(false);
+ // Optionally require that all cores on this node are active if param
'requireHealthyCores=true'
+ if (req.getParams().getBool(PARAM_REQUIRE_HEALTHY_CORES, false)) {
+ List<String> unhealthyCores = findUnhealthyCores(clusterState,
+ cores.getNodeConfig().getNodeName(),
+ cores.getAllCoreNames());
+ if (unhealthyCores.size() > 0) {
+ rsp.add(STATUS, FAILURE);
+ rsp.setException(new
SolrException(SolrException.ErrorCode.SERVICE_UNAVAILABLE,
+ "Replica(s) " + unhealthyCores + " are currently
initializing or recovering"));
+ return;
+ }
+ rsp.add("message", "All cores are healthy");
+ }
+
+ // All lights green, report healthy
+ rsp.add(STATUS, OK);
+ }
+
+ /**
+ * Find replicas DOWN or RECOVERING, or replicas in clusterstate that do not
exist on local node
+ * @param clusterState clusterstate from ZK
+ * @param nodeName this node name
+ * @param allCoreNames list of all core names on current node
+ * @return list of core names that are either DOWN ore RECOVERING on
'nodeName'
+ */
+ static List<String> findUnhealthyCores(ClusterState clusterState, String
nodeName, Collection<String> allCoreNames) {
+ return clusterState.getCollectionsMap().values().stream()
Review comment:
Why does this need to go to cluster state? This becomes a very expensive
method when you have a lot of collections because it fetches collection states
for all collections from ZK even if those collections have no replicas on the
current node. Imagine having 300 collections and doing 299 zk read operations
just for a health check if the node hosts replicas for one collection only.
I think this method should be rewritten to iterate over the CloudDescriptors
of all local cores and check for `hasRegistered == true` and `lastPublished ==
ACTIVE`. Those two should be sufficient for a health check functionality. We
don't even need to consult the cluster state.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
