[CARBONDATA-3665] Support TimeBased Cache expiration using Guava Cache #3580

Indhumathi27 · 2020-01-15T13:35:50Z

Why is this PR needed?

Currently, in Carbon, we follow LRU cache based mechanism. An least-recently used entry will be removed from the cache when it is full. There is no time-based cache expiration supported in carbon. In cloud, all vm's may not have enough memory to cache everything we could cache.
In that case, we can clear cache after a specified duration. This can be achieved by using cache libraries available.

One of the caching library is Guava Cache, which provides flexible and powerful caching features. Please refer GuavaCache for more info.

What changes were proposed in this PR?

Replaced LinkedHashMap with Guava Cache
Added Carbon property to allow user to specify cache expiration duration in minutes, to clear cache.
Newly added carbon property:
carbon.lru.cache.expiration.duration.in.minutes which takes long value.
For example:
carbon.lru.cache.expiration.duration.in.minutes="5" -> After 5 minutes, cache will be cleared.

Does this PR introduce any user interface change?

Yes. Added new property. Document is updated

Is any new testcase added?

Yes

CarbonDataQA1 · 2020-01-15T14:57:41Z

Build Success with Spark 2.3.4, Please check CI http://121.244.95.60:12545/job/ApacheCarbonPRBuilder2.3/1646/

CarbonDataQA1 · 2020-01-16T01:36:29Z

Build Failed with Spark 2.3.4, Please check CI http://121.244.95.60:12545/job/ApacheCarbonPRBuilder2.3/1647/

Indhumathi27 · 2020-01-16T11:59:10Z

retest this please

CarbonDataQA1 · 2020-01-16T13:25:27Z

Build Success with Spark 2.3.4, Please check CI http://121.244.95.60:12545/job/ApacheCarbonPRBuilder2.3/1663/

CarbonDataQA1 · 2020-01-17T03:22:46Z

Build Success with Spark 2.3.4, Please check CI http://121.244.95.60:12545/job/ApacheCarbonPRBuilder2.3/1672/

ajantha-bhat · 2020-01-19T01:43:54Z

core/src/main/java/org/apache/carbondata/core/cache/CarbonLRUCache.java

   * object
   */
-  private Map<String, Cacheable> lruCacheMap;
+  private Cache<String, Cacheable> lruCacheMap;


There are scenarios where cache of some table need to be maintained and other tables can be dropped. It will be good if we support table level expiry. and not clear all the cache when expire.

Drop meta cache DML is already supporting table level dropping , but it is manual work now. If this can be handled by guava, it is great

Currently, using guava cache, we can do only time-based or size-based eviction for all values(not specific to table level) loaded to CarbonLruCache.

@jackylk , @ravipesala : what do you think ?

@Indhumathi27 : I still think we must have entry level expiry, just removing all the cache after certain time is not so useful as cache need to be loaded again ?

check this guava issue for the same (here)
I just checked on internet and I found 2 alternatives (under apache license) that supports entry level expiry.
a. caffeine -- This is on top of guava as they support guava adaptor
b. expiringmap -- This is widely used by many companies in their project. see here. I guess we can use this one.

@Indhumathi27 i agree with @ajantha-bhat , may we need to keep option, or give choices to user and we should have implementations based on that like time based, size based.

@ajantha-bhat @akashrn5 ok. I will check those caching libraries and will raise new PR.

xuchuanyin · 2020-01-21T00:44:21Z

core/src/main/java/org/apache/carbondata/core/cache/CarbonLRUCache.java

+    String timeBasedExpiration = CarbonProperties.getInstance()
+        .getProperty(CarbonCommonConstants.CARBON_LRU_CACHE_EXPIRATION_DURATION_IN_MINUTES);
+    if (null != timeBasedExpiration) {
+      duration = Long.parseLong(timeBasedExpiration);


forget to validate the content?

Configuration is already validated in core/src/main/java/org/apache/carbondata/core/util/CarbonProperties.java. Please check

xuchuanyin · 2020-01-21T00:46:57Z

core/src/main/java/org/apache/carbondata/core/cache/CarbonLRUCache.java

-        new LinkedHashMap<String, Cacheable>(CarbonCommonConstants.DEFAULT_COLLECTION_SIZE, 1.0f,
-            true);
+        CacheBuilder.newBuilder().initialCapacity(CarbonCommonConstants.DEFAULT_COLLECTION_SIZE)
+            .expireAfterWrite(duration, TimeUnit.MINUTES).build();


expire after write or access?

besides, guava also support size based cache, will it be supported later?

Size-based eviction can be supported later if needed

xuchuanyin · 2020-01-21T00:49:03Z

core/src/main/java/org/apache/carbondata/core/cache/CarbonLRUCache.java

+    if (null != timeBasedExpiration) {
+      duration = Long.parseLong(timeBasedExpiration);
+    }
+    // initialise guava cache with time based expiration
    lruCacheMap =


better to optimize this variable's name -- lruCache is enough

xuchuanyin · 2020-01-21T00:57:49Z

core/src/main/java/org/apache/carbondata/core/cache/CarbonLRUCache.java

@@ -327,15 +309,15 @@ public Cacheable get(String key) {
   */
  public void clear() {
    synchronized (lruCacheMap) {
-      for (Cacheable cachebleObj : lruCacheMap.values()) {
+      for (Cacheable cachebleObj : lruCacheMap.asMap().values()) {


why not use method invalidateAll() ?

the above code is for invalidating Cacheable objects. For LRUCache, cleanUp method is used for clearing the cache.

CarbonDataQA1 · 2020-01-23T06:20:07Z

Build Failed with Spark 2.3.4, Please check CI http://121.244.95.60:12545/job/ApacheCarbonPRBuilder2.3/1752/

CarbonDataQA1 · 2020-01-23T07:43:41Z

Build Success with Spark 2.3.4, Please check CI http://121.244.95.60:12545/job/ApacheCarbonPRBuilder2.3/1754/

akashrn5 · 2020-01-26T15:21:31Z

core/src/main/java/org/apache/carbondata/core/util/CarbonProperties.java

+        CarbonCommonConstants.CARBON_LRU_CACHE_EXPIRATION_DURATION_IN_MINUTES_DEFAULT);
+    long duration;
+    try {
+      duration = Long.parseLong(


rename to carbonCacheExpireDuration

akashrn5 · 2020-01-26T15:24:04Z

core/src/main/java/org/apache/carbondata/core/util/CarbonProperties.java

+            .setProperty(CARBON_LRU_CACHE_EXPIRATION_DURATION_IN_MINUTES,
+                Long.toString(duration));
+      }
+    } catch (Exception e) {


In this method, only NumberFormatException will be thrown, i dont see any other type of exception. If not instead of generalised exception, use NumberFormatException.

akashrn5 · 2020-01-26T15:27:05Z

core/src/main/java/org/apache/carbondata/core/cache/CarbonLRUCache.java

   * object
   */
-  private Map<String, Cacheable> lruCacheMap;
+  private Cache<String, Cacheable> lruCache;


since now least recently used cache is changed to time based, does this variable name still makes sense? shall we give a better one to depict the time based cache implementation?

Indhumathi27 force-pushed the guava_cache branch 2 times, most recently from f677e3b to a3503ea Compare January 16, 2020 01:32

Indhumathi27 changed the title ~~[WIP] Support TimeBased Cache expiration using Guava Cache~~ [CARBONDATA-3665] Support TimeBased Cache expiration using Guava Cache Jan 16, 2020

Indhumathi27 changed the title ~~[CARBONDATA-3665] Support TimeBased Cache expiration using Guava Cache~~ [WIP][CARBONDATA-3665] Support TimeBased Cache expiration using Guava Cache Jan 16, 2020

Indhumathi27 force-pushed the guava_cache branch from a3503ea to 4008b44 Compare January 17, 2020 02:02

Indhumathi27 changed the title ~~[WIP][CARBONDATA-3665] Support TimeBased Cache expiration using Guava Cache~~ [CARBONDATA-3665] Support TimeBased Cache expiration using Guava Cache Jan 17, 2020

ajantha-bhat reviewed Jan 19, 2020

View reviewed changes

xuchuanyin reviewed Jan 21, 2020

View reviewed changes

Indhumathi27 force-pushed the guava_cache branch 2 times, most recently from 79ed635 to 754300d Compare January 23, 2020 06:18

Support TimeBased Cache expiration using Guava Cache

fd7c3a3

Indhumathi27 force-pushed the guava_cache branch from 754300d to fd7c3a3 Compare January 23, 2020 06:23

akashrn5 reviewed Jan 26, 2020

View reviewed changes

Indhumathi27 closed this Jan 28, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[CARBONDATA-3665] Support TimeBased Cache expiration using Guava Cache #3580

[CARBONDATA-3665] Support TimeBased Cache expiration using Guava Cache #3580

Indhumathi27 commented Jan 15, 2020 •

edited

CarbonDataQA1 commented Jan 15, 2020

CarbonDataQA1 commented Jan 16, 2020

Indhumathi27 commented Jan 16, 2020

CarbonDataQA1 commented Jan 16, 2020

CarbonDataQA1 commented Jan 17, 2020

ajantha-bhat Jan 19, 2020

Indhumathi27 Jan 22, 2020

ajantha-bhat Jan 27, 2020 •

edited

akashrn5 Jan 27, 2020

Indhumathi27 Jan 28, 2020

xuchuanyin Jan 21, 2020

Indhumathi27 Jan 22, 2020

xuchuanyin Jan 21, 2020

Indhumathi27 Jan 22, 2020

xuchuanyin Jan 21, 2020

Indhumathi27 Jan 23, 2020

xuchuanyin Jan 21, 2020

Indhumathi27 Jan 22, 2020

CarbonDataQA1 commented Jan 23, 2020

CarbonDataQA1 commented Jan 23, 2020

akashrn5 Jan 26, 2020

akashrn5 Jan 26, 2020

akashrn5 Jan 26, 2020

[CARBONDATA-3665] Support TimeBased Cache expiration using Guava Cache #3580

[CARBONDATA-3665] Support TimeBased Cache expiration using Guava Cache #3580

Conversation

Indhumathi27 commented Jan 15, 2020 • edited

Why is this PR needed?

What changes were proposed in this PR?

Does this PR introduce any user interface change?

Is any new testcase added?

CarbonDataQA1 commented Jan 15, 2020

CarbonDataQA1 commented Jan 16, 2020

Indhumathi27 commented Jan 16, 2020

CarbonDataQA1 commented Jan 16, 2020

CarbonDataQA1 commented Jan 17, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ajantha-bhat Jan 27, 2020 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

CarbonDataQA1 commented Jan 23, 2020

CarbonDataQA1 commented Jan 23, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Indhumathi27 commented Jan 15, 2020 •

edited

ajantha-bhat Jan 27, 2020 •

edited