add bulk tuning for translog (async mode) and fast replica change

This commit is contained in:
Jörg Prante 2021-05-04 18:32:57 +02:00
parent a691ace2d5
commit 97f460ae98
4 changed files with 29 additions and 5 deletions

View file

@ -120,9 +120,23 @@ public abstract class AbstractAdminClient extends AbstractBasicClient implements
return this;
}
logger.info("update replica level for " + indexDefinition + " to " + indexDefinition.getReplicaCount());
int currentReplicaLevel = getReplicaLevel(indexDefinition);
logger.info("current replica level for " + indexDefinition + " is " + currentReplicaLevel);
if (currentReplicaLevel < indexDefinition.getReplicaCount()) {
putClusterSetting("cluster.routing.allocation.node_concurrent_recoveries", "5", 30L, TimeUnit.SECONDS);
putClusterSetting("indices.recovery.max_bytes_per_sec", "2gb", 30L, TimeUnit.SECONDS);
logger.info("recovery boost activated");
}
updateIndexSetting(indexDefinition.getFullIndexName(), "number_of_replicas",
indexDefinition.getReplicaCount(), 30L, TimeUnit.SECONDS);
waitForHealthyCluster();
if (currentReplicaLevel < indexDefinition.getReplicaCount()) {
// 2 = default setting
putClusterSetting("cluster.routing.allocation.node_concurrent_recoveries", "2", 30L, TimeUnit.SECONDS);
// 20m = default value
putClusterSetting("indices.recovery.max_bytes_per_sec", "40mb", 30L, TimeUnit.SECONDS);
logger.info("recovery boost deactivated");
}
return this;
}

View file

@ -9,6 +9,7 @@ import org.elasticsearch.action.admin.cluster.health.ClusterHealthRequest;
import org.elasticsearch.action.admin.cluster.health.ClusterHealthResponse;
import org.elasticsearch.action.admin.cluster.settings.ClusterUpdateSettingsAction;
import org.elasticsearch.action.admin.cluster.settings.ClusterUpdateSettingsRequest;
import org.elasticsearch.action.admin.cluster.settings.ClusterUpdateSettingsResponse;
import org.elasticsearch.action.admin.cluster.state.ClusterStateAction;
import org.elasticsearch.action.admin.cluster.state.ClusterStateRequest;
import org.elasticsearch.action.admin.cluster.state.ClusterStateResponse;
@ -118,7 +119,11 @@ public abstract class AbstractBasicClient implements BasicClient {
updateSettingsBuilder.put(key, value.toString());
ClusterUpdateSettingsRequest updateSettingsRequest = new ClusterUpdateSettingsRequest();
updateSettingsRequest.transientSettings(updateSettingsBuilder).timeout(toTimeValue(timeout, timeUnit));
client.execute(ClusterUpdateSettingsAction.INSTANCE, updateSettingsRequest).actionGet();
ClusterUpdateSettingsResponse clusterUpdateSettingsResponse =
client.execute(ClusterUpdateSettingsAction.INSTANCE, updateSettingsRequest).actionGet();
if (clusterUpdateSettingsResponse.isAcknowledged()) {
logger.info("cluster update of " + key + " to " + value + " acknowledged");
}
}
@Override
@ -139,7 +144,8 @@ public abstract class AbstractBasicClient implements BasicClient {
.waitForStatus(status);
ClusterHealthResponse healthResponse =
client.execute(ClusterHealthAction.INSTANCE, clusterHealthRequest).actionGet();
if (healthResponse != null && healthResponse.isTimedOut()) {
logger.info("got cluster status " + healthResponse.getStatus().name());
if (healthResponse.isTimedOut()) {
String message = "timeout, cluster state is " + healthResponse.getStatus().name() + " and not " + status.name();
logger.error(message);
throw new IllegalStateException(message);

View file

@ -141,8 +141,10 @@ public abstract class AbstractBulkClient extends AbstractBasicClient implements
int interval = indexDefinition.getStartBulkRefreshSeconds();
if (interval != 0) {
logger.info("starting bulk on " + indexName + " with new refresh interval " + interval);
updateIndexSetting(indexName, "refresh_interval",
interval >=0 ? interval + "s" : interval, 30L, TimeUnit.SECONDS);
updateIndexSetting(indexName,
"refresh_interval", interval >=0 ? interval + "s" : interval, 30L, TimeUnit.SECONDS);
updateIndexSetting(indexName,
"index.translog.durability", "async", 30L, TimeUnit.SECONDS);
} else {
logger.warn("ignoring starting bulk on " + indexName + " with refresh interval " + interval);
}
@ -168,6 +170,8 @@ public abstract class AbstractBulkClient extends AbstractBasicClient implements
logger.info("stopping bulk on " + indexName + " with new refresh interval " + interval);
updateIndexSetting(indexName, "refresh_interval",
interval >= 0 ? interval + "s" : interval, 30L, TimeUnit.SECONDS);
updateIndexSetting(indexName,
"index.translog.durability", "request", 30L, TimeUnit.SECONDS);
} else {
logger.warn("ignoring stopping bulk on " + indexName + " with refresh interval " + interval);
}

View file

@ -1,6 +1,6 @@
group = org.xbib
name = elx
version = 7.10.2.6
version = 7.10.2.7
gradle.wrapper.version = 6.6.1
xbib-metrics.version = 2.1.0