[
https://issues.apache.org/jira/browse/GEODE-8130?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17107413#comment-17107413
]
ASF GitHub Bot commented on GEODE-8130:
---------------------------------------
ringles commented on a change in pull request #5067:
URL: https://github.com/apache/geode/pull/5067#discussion_r425234480
##########
File path:
geode-redis/src/main/java/org/apache/geode/redis/internal/KeyRegistrar.java
##########
@@ -14,56 +14,147 @@
*/
package org.apache.geode.redis.internal;
+import java.io.DataInput;
+import java.io.DataOutput;
+import java.io.IOException;
import java.util.Map;
import java.util.Set;
+import org.apache.geode.DataSerializer;
+import org.apache.geode.InvalidDeltaException;
import org.apache.geode.cache.Region;
public class KeyRegistrar {
- private Region<String, RedisDataType> redisMetaRegion;
+ private Region<ByteArrayWrapper, RedisData> redisDataRegion;
- public KeyRegistrar(Region<String, RedisDataType> redisMetaRegion) {
- this.redisMetaRegion = redisMetaRegion;
+ public KeyRegistrar(Region<ByteArrayWrapper, RedisData> redisDataRegion) {
+ this.redisDataRegion = redisDataRegion;
}
/**
* Checks if the givenKey is associated with the passed data type. If an
entry doesn't exist,
* store the key:datatype association in the metadataRegion
*/
public void register(ByteArrayWrapper key, RedisDataType type) {
- RedisDataType existingType =
this.redisMetaRegion.putIfAbsent(key.toString(), type);
- if (!isValidDataType(existingType, type)) {
- throwDataTypeException(key, existingType);
+ RedisData existingValue = this.redisDataRegion.putIfAbsent(key,
transformType(type));
+ if (!isValidDataType(existingValue, type)) {
+ throwDataTypeException(key, existingValue);
+ }
+ }
+
+ /**
+ * This class should go away once all data types implement RedisData.
Review comment:
Probably a TODO here, so it's more obvious.
##########
File path:
geode-redis/src/main/java/org/apache/geode/redis/internal/GeodeRedisServer.java
##########
@@ -526,20 +499,28 @@ private void registerLockServiceMBean() {
}
private void checkForRegions() {
- Collection<Entry<String, RedisDataType>> entrySet =
keyRegistrar.keyInfos();
- for (Entry<String, RedisDataType> entry : entrySet) {
- String regionName = entry.getKey();
- RedisDataType type = entry.getValue();
- Region<?, ?> newRegion = cache.getRegion(regionName);
- if (newRegion == null && type != RedisDataType.REDIS_STRING && type !=
RedisDataType.REDIS_HLL
- && type != RedisDataType.REDIS_PROTECTED) {
- try {
- regionCache
-
.createRemoteRegionReferenceLocally(Coder.stringToByteArrayWrapper(regionName),
type);
- } catch (Exception e) {
- if (logger.errorEnabled()) {
- logger.error(e);
- }
+ Collection<Entry<ByteArrayWrapper, RedisData>> entrySet =
keyRegistrar.keyInfos();
+ for (Entry<ByteArrayWrapper, RedisData> entry : entrySet) {
+ ByteArrayWrapper key = entry.getKey();
+ RedisDataType type = entry.getValue().getType();
+ if (!regionCache.typeUsesDynamicRegions(type)) {
Review comment:
Should "regionCache" be renamed "regionProvider"? Line 506 talks about a
'regionCache', and then 509 has a 'cache'. I can see someone looking over the
code wondering, "why two caches"?
----------------------------------------------------------------
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:
[email protected]
> Change the way redis uses geode regions to save memory and fix race conditions
> ------------------------------------------------------------------------------
>
> Key: GEODE-8130
> URL: https://issues.apache.org/jira/browse/GEODE-8130
> Project: Geode
> Issue Type: Improvement
> Components: redis
> Reporter: Darrel Schneider
> Priority: Major
>
> Currently the geode redis implementation uses a replicate region to hold
> every redis key as the region key (as a String object) and the region value
> is an enum describing the type. Another partitioned regioni is used to also
> store the redis key as the region key (as a ByteArrayWrapper object) and the
> region value is an object that is basically a Set or HashMap.
> This uses extra memory, the replicate does not scale well, is slower
> (updating two regions is slower than one region), and trying to update two
> regions to hold the same logical redis "key" has race conditions which can
> lead to inconsistent data.
> The solution is to have a single partitioned region that can hold multiple
> types of data (sets, hashes, strings, lists, etc).
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
