zabetak commented on code in PR #5764:
URL: https://github.com/apache/hive/pull/5764#discussion_r2041704820
##########
standalone-metastore/metastore-common/src/main/java/org/apache/hadoop/hive/metastore/conf/MetastoreConf.java:
##########
@@ -700,6 +700,12 @@ public enum ConfVars {
"configured with embedded metastore. To get optimal performance,
set config to meet the following condition\n"+
"(2 * pool_size * metastore_instances + 2 * pool_size *
HS2_instances_with_embedded_metastore) = \n" +
"(2 * physical_core_count + hard_disk_count)."),
+
CONNECTION_POOLING_MAX_SECONDARY_CONNECTIONS("datanucleus.connectionPool.maxSecondaryPoolSize",
+ "datanucleus.connectionPool.maxSecondaryPoolSize", 2,
+ "Specify the maximum number of connections in the
objectstore-secondary connection pool.\n" +
+ "Same as datanucleus.connectionPool.maxPoolSize, but for
the objectstore-secondary pool. \n" +
+ "Note that this is not a valid datanucleus configuration,
this is used only in Hive.\n" +
Review Comment:
The comment it is not a valid configuration is a bit confusing. Note that
`datanucleus.connectionPool.maxPoolSize` is not passed directly to datanucleus
and in fact it is only used by Hive. It may be better to not mention anything.
##########
standalone-metastore/metastore-server/src/main/java/org/apache/hadoop/hive/metastore/PersistenceManagerProvider.java:
##########
@@ -315,9 +315,12 @@ private static PersistenceManagerFactory
initPMF(Configuration conf, boolean for
databaseProduct = DatabaseProduct.determineDatabaseProduct(ds, conf);
// The secondary connection factory is used for schema generation, and
for value generation operations.
// We should use a different pool for the secondary connection factory
to avoid resource starvation.
- // Since DataNucleus uses locks for schema generation and value
generation, 2 connections should be sufficient.
+ // DataNucleus uses locks for schema generation and value generation,
under normal circumstances 2 connections
+ // should be sufficient. However, as found in HIVE-28839 in certain
situations connection starvation may happen,
+ // so we need to make this configurable until a final fix is not
available.
Review Comment:
As far as I understand the intention is to keep this config around even
after fixing HIVE-28839. If my understanding is correct then it's better to
remove the comment "However...is not available".
If it is really meant to be a temporary config till we merge HIVE-28839 then
it's fine to leave things as is. Maybe just fix the typo "fix is not available"
-> "fix is available".
##########
standalone-metastore/metastore-common/src/main/java/org/apache/hadoop/hive/metastore/conf/MetastoreConf.java:
##########
@@ -700,6 +700,12 @@ public enum ConfVars {
"configured with embedded metastore. To get optimal performance,
set config to meet the following condition\n"+
"(2 * pool_size * metastore_instances + 2 * pool_size *
HS2_instances_with_embedded_metastore) = \n" +
"(2 * physical_core_count + hard_disk_count)."),
+
CONNECTION_POOLING_MAX_SECONDARY_CONNECTIONS("datanucleus.connectionPool.maxSecondaryPoolSize",
Review Comment:
I would prefer slightly more a name like
`datanucleus.connectionPool.secondary.maxPoolSize` or
`datanucleus.connectionPool.nontx.maxPoolSize` but don't feel very strong about
it.
The datanucleus prefix is somewhat misleading since we are not propagating
the property to datanucleus but given the current situation and the other props
don't have a better suggestion so we can leave with it.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]