### What changes were proposed in this pull request?
Cleans up the pooled send buffers and push tasks if the SendBufferPool has been idle for more than
`celeborn.client.push.sendbufferpool.expireTimeout`.
### Why are the changes needed?
Before this PR the SendBufferPool will cache the send buffers and push tasks forever. If they are large
and will not be reused in the future, it wastes memory and causes GC.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Passes GA and manual tests.
Closes#1735 from waitinfuture/812-1.
Authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
Fix some typos and grammar
### Why are the changes needed?
Ditto
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
manually test
Closes#1733 from onebox-li/fix-typo.
Authored-by: onebox-li <lyh-36@163.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
As title.
### Why are the changes needed?
Timeout of ```RpcEndpointRef.ask``` is controlled by ```celeborn.rpc.askTimeout```,
so we also need to increase ```celeborn.rpc.askTimeout``` to extend the timeout of commit files.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Passes GA and manual test.
Closes#1725 from waitinfuture/803-fu.
Authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
As title.
### Why are the changes needed?
In 0.2.1-incubating, commit files default timeout is ```NETWORK_TIMEOUT```, which is 240s.
It's more reasonable because commit files costs relatively long time. In my testing with tough disks,
30s timeout with 2 retires is not enough.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Passes GA and manual test.
Closes#1724 from waitinfuture/803.
Authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
…up client
### What changes were proposed in this pull request?
Add heartbeat from client to lifecycle manager. In this PR heartbeat request contains local shuffle ids from
client, lifecycle manager checks with it's local set and returns ids it doesn't know. Upon receiving response,
client calls ```unregisterShuffle``` for cleanup.
### Why are the changes needed?
Before this PR, client side ```unregisterShuffle``` is never called. When running TPCDS 3T with spark thriftserver
without DRA, I found the Executor's heap contains 1.6 million PartitionLocation objects (and StorageInfo):
After this PR, the number of PartitionLocation objects decreases to 275 thousands
This heartbeat can be extended in the future for other purposes, i.e. reporting client's metrics.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Passes GA and manual test.
Closes#1719 from waitinfuture/798.
Authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
…Flight.total```
### What changes were proposed in this pull request?
Refer to https://github.com/apache/incubator-celeborn/pull/1720#discussion_r1265092164
### Why are the changes needed?
ditto
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Passes GA.
Closes#1723 from waitinfuture/799-fu.
Authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
As title.
### Why are the changes needed?
In case where worker instances is very large, say 1000, then before this PR total memory consumed
by inflight requests is 64K * 1000 * ```celeborn.client.push.maxReqsInFlight(16)``` = 1G. This PR
limits total inflight push requests, as 0.2.1-incubating does.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Passes GA and manual test.
Closes#1720 from waitinfuture/799.
Lead-authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Co-authored-by: Cheng Pan <pan3793@gmail.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
1. Decrease writeTime metric sampling frequency to improve perf
2. Set default value of ```celeborn.<module>.push.timeoutCheck.threads``` and ```celeborn.<module>.fetch.timeoutCheck.threads``` to 4
### Why are the changes needed?
Following are test cases
case 1: ```spark.sparkContext.parallelize(1 to 8000, 8000).flatMap( _ => (1 to 15000000).iterator.map(num => num)).repartition(8000).count``` // shuffle 1.1T data
case 2: ```spark.sparkContext.parallelize(1 to 8000, 8000).flatMap( _ => (1 to 30000000).iterator.map(num => num)).repartition(8000).count``` // shuffle 2.2T data
Following are e2e time of shuffle write stage
||Sort pusher before|Sort pusher after|Hash pusher before|Hash pusher after|
|----|----|----|----|-----|
|case1|4.4min|4.1min|4.4min|3.9min|
|case2|9.1min|8.4min|9.7min|8.5min|
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Passes GA and manual test.
Closes#1718 from waitinfuture/797.
Authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
Add a configuration `celeborn.worker.shuffle.partitionSplit.max` to ensure that, in soft mode, individual partition files are limited to a size smaller than the configured value
### Why are the changes needed?
In soft mode, there may be situations where individual partition files are exceptionally large, which can result in excessively long sort times in skewed scenarios.
### Does this PR introduce _any_ user-facing change?
`celeborn.worker.shuffle.partitionSplit.max` defalut value 2g
### How was this patch tested?
none
Closes#1701 from JQ-Cao/785.
Authored-by: caojiaqing <caojiaqing@bilibili.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
This PR changes default values of the following configs:
|config|previous default value|new default value|
|----|----|----|
|celeborn.worker.flusher.threads|2|16|
|celeborn.worker.flusher.ssd.threads|8|16|
### Why are the changes needed?
If disk type is not specified, ```celeborn.worker.flusher.threads``` will be used. Recently many users
use SSD for Celeborn workers without specifying disk type, and 2 flush threads is far from leveraging the power of SSD.
### Does this PR introduce _any_ user-facing change?
Yes, default configs are changed.
### How was this patch tested?
Passes GA.
Closes#1703 from waitinfuture/786.
Authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
Now slots is not a bottleneck, change SPARK_SHUFFLE_FORCE_FALLBACK_PARTITION_THRESHOLD default value to Int.MaxValue.
### Why are the changes needed?
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Closes#1695 from AngersZhuuuu/CELEBORN-780.
Authored-by: Angerszhuuuu <angers.zhu@gmail.com>
Signed-off-by: Angerszhuuuu <angers.zhu@gmail.com>
### What changes were proposed in this pull request?
Pullout hardcoded `celeborn.rpc.dispatcher.numThreads` to `CelebornConf` and rename it to `celeborn.rpc.dispatcher.threads` to align with existing configuration style
### Why are the changes needed?
Pullout inline configuration to `CelebornConf`, and expose it in configuration docs
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Pass GA.
Closes#1684 from pan3793/CELEBORN-774.
Authored-by: Cheng Pan <chengpan@apache.org>
Signed-off-by: Cheng Pan <chengpan@apache.org>
…Flight to 16
### What changes were proposed in this pull request?
Change default value of celeborn.client.push.maxReqsInFlight to 16.
### Why are the changes needed?
Previous value 4 is too small, 16 is more reasonable.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Pass GA.
Closes#1683 from waitinfuture/769.
Authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
Make Celeborn leader clean expired app dirs on HDFS when an application is Lost.
### Why are the changes needed?
If Celeborn is working on HDFS, the storage manager starts and cleans expired app directories, and the newly created worker will want to delete any unknown app directories.
This will cause using app directories to be deleted unexpectedly.
### Does this PR introduce _any_ user-facing change?
NO.
### How was this patch tested?
UT and cluster.
Closes#1678 from FMX/CELEBORN-764.
Lead-authored-by: mingji <fengmingxiao.fmx@alibaba-inc.com>
Co-authored-by: Cheng Pan <pan3793@gmail.com>
Co-authored-by: Ethan Feng <fengmingxiao.fmx@alibaba-inc.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
…memory allocator
### What changes were proposed in this pull request?
Changes the following configs' default values
| config | previous value | current value |
| ------------- | ------------- | ------------- |
| celeborn.network.memory.allocator.share | false | true |
| celeborn.client.shuffle.batchHandleChangePartition.enabled | false | true |
| celeborn.client.shuffle.batchHandleCommitPartition.enabled | false | true |
### Why are the changes needed?
In my test, when graceful shutdown is enabled but ```celeborn.client.shuffle.batchHandleChangePartition.enabled``` and ```celeborn.client.shuffle.batchHandleCommitPartition.enabled``` disabled, the worker takes much longer to stop than the two configs enabled.
In another test where worker size is quite small(2 cores 4 G) and replication is on, if shared allocator is disabled, the netty's onTrim fails to release memory, and further causes push data timeout.
### Does this PR introduce _any_ user-facing change?
No, these conifgs are introduces from 0.3.0.
### How was this patch tested?
Passes GA.
Closes#1682 from waitinfuture/768.
Authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
As title
### Why are the changes needed?
To clarify the usage of conf `celeborn.client.spark.push.sort.memory.threshold`
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Pass GA
Closes#1680 from cfmcgrady/docs.
Authored-by: Fu Chen <cfmcgrady@gmail.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
Rename remain rss related class name and filenames etc...
### Why are the changes needed?
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Closes#1664 from AngersZhuuuu/CELEBORN-751.
Authored-by: Angerszhuuuu <angers.zhu@gmail.com>
Signed-off-by: Angerszhuuuu <angers.zhu@gmail.com>
### What changes were proposed in this pull request?
as title
### Why are the changes needed?
mention configuration behavior change in migration guide
### Does this PR introduce _any_ user-facing change?
Yes, the migration guide is updated
### How was this patch tested?
review
Closes#1673 from pan3793/CELEBORN-637-followup.
Authored-by: Cheng Pan <chengpan@apache.org>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
Support to decide whether to compress shuffle data through configuration.
### Why are the changes needed?
Currently, Celeborn compresses all shuffle data, but for example, the shuffle data of Gluten has already been compressed. In this case, no additional compression is required. Therefore, configuration needs to be provided for users to decide whether to use Celeborn’s compression according to the actual situation.
### Does this PR introduce _any_ user-facing change?
no.
Closes#1669 from kerwin-zk/celeborn-755.
Authored-by: xiyu.zk <xiyu.zk@alibaba-inc.com>
Signed-off-by: Cheng Pan <chengpan@apache.org>
### What changes were proposed in this pull request?
Provide a new SparkShuffleManager to replace RssShuffleManager in the future
### Why are the changes needed?
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Closes#1667 from AngersZhuuuu/CELEBORN-754.
Lead-authored-by: Angerszhuuuu <angers.zhu@gmail.com>
Co-authored-by: Cheng Pan <pan3793@gmail.com>
Signed-off-by: Cheng Pan <chengpan@apache.org>
### What changes were proposed in this pull request?
As title
### Why are the changes needed?
In order to distinguish it from the existing master/worker, refactor data replication terminology to 'primary/replica' for improved clarity and inclusivity in the codebase
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
existing tests
Closes#1639 from cfmcgrady/primary-replica.
Lead-authored-by: Fu Chen <cfmcgrady@gmail.com>
Co-authored-by: Cheng Pan <chengpan@apache.org>
Signed-off-by: Cheng Pan <chengpan@apache.org>
### What changes were proposed in this pull request?
This pull PR is an integral component of #1639 . It primarily focuses on updating configuration settings and metrics terminology, while ensuring compatibility with older client versions by refraining from introducing changes related to RPC.
### Why are the changes needed?
In order to distinguish it from the existing master/worker, refactor data replication terminology to 'primary/replica' for improved clarity and inclusivity in the codebase
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
existing tests.
Closes#1650 from cfmcgrady/primary-replica-metrics.
Authored-by: Fu Chen <cfmcgrady@gmail.com>
Signed-off-by: Cheng Pan <chengpan@apache.org>
### What changes were proposed in this pull request?
Refine the congestion relevant code/log/comments
### Why are the changes needed?
ditto
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
manually test
Closes#1637 from onebox-li/improve-congestion.
Authored-by: onebox-li <lyh-36@163.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
Unify all blacklist related code and comment
### Why are the changes needed?
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Closes#1638 from AngersZhuuuu/CELEBORN-666-FOLLOWUP.
Authored-by: Angerszhuuuu <angers.zhu@gmail.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
…nse with lower versions
### What changes were proposed in this pull request?
The master side will check HeartbeatFromApplication's reply field. if reply is true then it replies HeartbeatFromApplicationResponse otherwise OneWayMessageResponse.
The reply field is default false before the version 0.2.1, so master can be compatible with older client version
### Why are the changes needed?
Before the version `0.2.1`, the response of HeartbeatFromApplication is` OneWayMessageResponse`, but from `0.3.0`, the response of HeartbeatFromApplication is modified to `HeartbeatFromApplicationResponse`.
if the version of `client side `is `0.2.1` and the version of `server side is 0.3.0`, the `compatiblity issue `will occur.
The following compatiblity error will be printted.
``` java
java.io.InvalidObjectException: enum constant HEARTBEAT_FROM_APPLICATION_RESPONSE does not exist in class org.apache.celeborn.common.protocol.MessageType
at java.io.ObjectInputStream.readEnum(ObjectInputStream.java:2157) ~[?:1.8.0_362]
at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1662) ~[?:1.8.0_362]
at java.io.ObjectInputStream.defaultReadFields(ObjectInputStream.java:2430) ~[?:1.8.0_362]
at java.io.ObjectInputStream.readSerialData(ObjectInputStream.java:2354) ~[?:1.8.0_362]
at java.io.ObjectInputStream.readOrdinaryObject(ObjectInputStream.java:2212) ~[?:1.8.0_362]
at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1668) ~[?:1.8.0_362]
at java.io.ObjectInputStream.readObject(ObjectInputStream.java:502) ~[?:1.8.0_362]
at java.io.ObjectInputStream.readObject(ObjectInputStream.java:460) ~[?:1.8.0_362]
at org.apache.celeborn.common.serializer.JavaDeserializationStream.readObject(JavaSerializer.scala:76) ~[celeborn-client-spark-3-shaded_2.12-0.2.1-incubating.jar:?]
```
``` java
Caused by: java.lang.ClassCastException: Cannot cast org.apache.celeborn.common.protocol.message.ControlMessages$HeartbeatFromApplicationResponse to org.apache.celeborn.common.protocol.message.ControlMessages$OneWayMessageResponse$
at java.lang.Class.cast(Class.java:3369) ~[?:1.8.0_362]
at scala.concurrent.Future.$anonfun$mapTo$1(Future.scala:500) ~[scala-library-2.12.15.jar:?]
at scala.util.Success.$anonfun$map$1(Try.scala:255) ~[scala-library-2.12.15.jar:?]
at scala.util.Success.map(Try.scala:213) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.Future.$anonfun$map$1(Future.scala:292) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.impl.Promise.liftedTree1$1(Promise.scala:33) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.impl.Promise.$anonfun$transform$1(Promise.scala:33) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.impl.CallbackRunnable.run(Promise.scala:64) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.BatchingExecutor$Batch.processBatch$1(BatchingExecutor.scala:67) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.BatchingExecutor$Batch.$anonfun$run$1(BatchingExecutor.scala:82) ~[scala-library-2.12.15.jar:?]
at scala.runtime.java8.JFunction0$mcV$sp.apply(JFunction0$mcV$sp.java:23) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.BlockContext$.withBlockContext(BlockContext.scala:85) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.BatchingExecutor$Batch.run(BatchingExecutor.scala:59) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.Future$InternalCallbackExecutor$.unbatchedExecute(Future.scala:875) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.BatchingExecutor.execute(BatchingExecutor.scala:110) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.BatchingExecutor.execute$(BatchingExecutor.scala:107) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.Future$InternalCallbackExecutor$.execute(Future.scala:873) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.impl.CallbackRunnable.executeWithValue(Promise.scala:72) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.impl.Promise$DefaultPromise.$anonfun$tryComplete$1(Promise.scala:288) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.impl.Promise$DefaultPromise.$anonfun$tryComplete$1$adapted(Promise.scala:288) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.impl.Promise$DefaultPromise.tryComplete(Promise.scala:288) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.Promise.trySuccess(Promise.scala:94) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.Promise.trySuccess$(Promise.scala:94) ~[scala-library-2.12.15.jar:?]
at scala.concurrent.impl.Promise$DefaultPromise.trySuccess(Promise.scala:187) ~[scala-library-2.12.15.jar:?]
at org.apache.celeborn.common.rpc.netty.NettyRpcEnv.onSuccess$1(NettyRpcEnv.scala:218) ~[celeborn-client-spark-3-shaded_2.12-0.2.1-incubating.jar:?]
```
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
The pr is tested manually and the testing process is as follows:
1. server side is deploy using the code of latest branch-0.3.
2. spark client is deploy the version of 0.2.1, then run spark-sql to execute 3 tpcds queries( query1.sql/querey2/quere3.sql whose datasize is 1T), finnally verify that the queries are executed successfully and no above compatiblity error printted
3. spark client is deploy the version of 0.3.0, then run spark-sql to execute 3 tpcds queries( query1.sql/querey2/quere3.sql whose datasize is 1T), finnally verify that the queries are executed successfully and no above compatiblity error printted
This patch had conflicts when merged, resolved by
Committer: Cheng Pan <chengpan@apache.org>
Closes#1635 from zhongqiangczq/heartbeat2.
Authored-by: zhongqiang.czq <zhongqiang.czq@alibaba-inc.com>
Signed-off-by: Cheng Pan <chengpan@apache.org>
### What changes were proposed in this pull request?
Unify exclude and blacklist related configuration
### Why are the changes needed?
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Closes#1633 from AngersZhuuuu/CELEBORN-666-NEW.
Authored-by: Angerszhuuuu <angers.zhu@gmail.com>
Signed-off-by: Angerszhuuuu <angers.zhu@gmail.com>
### What changes were proposed in this pull request?
This PR batches revive requests and periodically send to LifecycleManager to reduce number or RPC requests.
To be more detailed. This PR changes Revive message to support multiple unique partitions, and also passes a set unique mapIds for checking MapEnd. Each time ShuffleClientImpl wants to revive, it adds a ReviveRquest to ReviveManager and wait for result. ReviveManager batches revive requests and periodically send to LifecycleManager (deduplicated by partitionId). LifecycleManager constructs ChangeLocationsCallContext and after all locations are notified, it replies to ShuffleClientImpl.
### Why are the changes needed?
In my test 3T TPCDS q23a with 3 Celeborn workers, when kill a worker, the LifecycleManger will receive 4.8w Revive requests:
```
[emr-usermaster-1-1 logs]$ cat spark-emr-user-org.apache.spark.sql.hive.thriftserver.HiveThriftServer2-1-master-1-1.c-fa08904e94c028d1.out.1 |grep -i revive |wc -l
64364
```
After this PR, number of ReviveBatch requests reduces to 708:
```
[emr-usermaster-1-1 logs]$ cat spark-emr-user-org.apache.spark.sql.hive.thriftserver.HiveThriftServer2-1-master-1-1.c-fa08904e94c028d1.out |grep -i revive |wc -l
2573
```
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Manual test. I have tested:
1. Disable graceful shutdown, kill one worker, job succeeds
2. Disable graceful shutdown, kill two workers successively, job fails as expected
3. Enable graceful shutdown, restart two workers successively, job succeeds
4. Enable graceful shutdown, restart two workers successively, then kill the third one, job succeeds
Closes#1588 from waitinfuture/656-2.
Lead-authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Co-authored-by: Keyong Zhou <zhouky@apache.org>
Co-authored-by: Keyong Zhou <waitinfuture@gmail.com>
Signed-off-by: Shuang <lvshuang.tb@gmail.com>
### What changes were proposed in this pull request?
1. Celeborn supports storage type selection. HDD, SSD, and HDFS are available for now.
2. Add new buffer size for HDFS file writers.
3. Worker support empty working dirs.
### Why are the changes needed?
Support HDFS only scenario.
### Does this PR introduce _any_ user-facing change?
NO.
### How was this patch tested?
UT and cluster.
Closes#1619 from FMX/CELEBORN-568.
Lead-authored-by: mingji <fengmingxiao.fmx@alibaba-inc.com>
Co-authored-by: Ethan Feng <fengmingxiao.fmx@alibaba-inc.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
Celeborn generate hadoop configuration should respect Celeborn conf
### Why are the changes needed?
In spark client side we should write like `spark.celeborn.hadoop.xxx.xx`
In server side we should write like `celeborn.hadoop.xxx.xxx`
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Closes#1629 from AngersZhuuuu/CELEBORN-719.
Authored-by: Angerszhuuuu <angers.zhu@gmail.com>
Signed-off-by: Angerszhuuuu <angers.zhu@gmail.com>
### What changes were proposed in this pull request?
This PR aims to make network local address binding support both IP and FQDN strategy.
Additional, it refactors the `ShuffleClientImpl#genAddressPair`, from `${hostAndPort}-${hostAndPort}` to `Pair<String, String>`, which works properly when using IP but may not on FQDN because FQDN may contain `-`
### Why are the changes needed?
Currently, when the bind hostname is not set explicitly, Celeborn will find the first non-loopback address and always uses the IP to bind, this is not suitable for K8s cases, as the STS has a stable FQDN but Pod IP will be changed once Pod restarting.
For `ShuffleClientImpl#genAddressPair`, it must be changed otherwise may cause
```
java.lang.RuntimeException: org.apache.spark.SparkException: Job aborted due to stage failure: Task 11657 in stage 0.0 failed 4 times, most recent failure: Lost task 11657.3 in stage 0.0 (TID 12747) (10.153.253.198 executor 157): java.lang.ArrayIndexOutOfBoundsException: 1
at org.apache.celeborn.client.ShuffleClientImpl.doPushMergedData(ShuffleClientImpl.java:874)
at org.apache.celeborn.client.ShuffleClientImpl.pushOrMergeData(ShuffleClientImpl.java:735)
at org.apache.celeborn.client.ShuffleClientImpl.mergeData(ShuffleClientImpl.java:827)
at org.apache.spark.shuffle.celeborn.SortBasedPusher.pushData(SortBasedPusher.java:140)
at org.apache.spark.shuffle.celeborn.SortBasedPusher.insertRecord(SortBasedPusher.java:192)
at org.apache.spark.shuffle.celeborn.SortBasedShuffleWriter.fastWrite0(SortBasedShuffleWriter.java:192)
at org.apache.spark.shuffle.celeborn.SortBasedShuffleWriter.write(SortBasedShuffleWriter.java:145)
at org.apache.spark.shuffle.ShuffleWriteProcessor.write(ShuffleWriteProcessor.scala:59)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:99)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:52)
at org.apache.spark.scheduler.Task.run(Task.scala:136)
at org.apache.spark.executor.Executor$TaskRunner.$anonfun$run$3(Executor.scala:548)
at org.apache.spark.util.Utils$.tryWithSafeFinally(Utils.scala:1508)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:551)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
at java.lang.Thread.run(Thread.java:750)
```
### Does this PR introduce _any_ user-facing change?
Yes, a new configuration `celeborn.network.bind.preferIpAddress` is introduced, and the default value is `true` to preserve the existing behavior.
### How was this patch tested?
Manually testing with `celeborn.network.bind.preferIpAddress=false`
```
Server: 10.178.96.64
Address: 10.178.96.64#53
Name: celeborn-master-0.celeborn-master-svc.spark.svc.cluster.local
Address: 10.153.143.252
Server: 10.178.96.64
Address: 10.178.96.64#53
Name: celeborn-master-1.celeborn-master-svc.spark.svc.cluster.local
Address: 10.153.173.94
Server: 10.178.96.64
Address: 10.178.96.64#53
Name: celeborn-master-2.celeborn-master-svc.spark.svc.cluster.local
Address: 10.153.149.42
starting org.apache.celeborn.service.deploy.worker.Worker, logging to /opt/celeborn/logs/celeborn--org.apache.celeborn.service.deploy.worker.Worker-1-celeborn-worker-4.out
2023-06-25 23:49:52 [INFO] [main] org.apache.celeborn.common.rpc.netty.Dispatcher#51 - Dispatcher numThreads: 4
2023-06-25 23:49:52 [INFO] [main] org.apache.celeborn.common.network.client.TransportClientFactory#91 - mode NIO threads 64
2023-06-25 23:49:52 [INFO] [main] org.apache.celeborn.common.rpc.netty.NettyRpcEnvFactory#51 - Starting RPC Server [WorkerSys] on celeborn-worker-4.celeborn-worker-svc.spark.svc.cluster.local:0 with advisor endpoint celeborn-worker-4.celeborn-worker-svc.spark.svc.cluster.local:0
2023-06-25 23:49:52 [INFO] [main] org.apache.celeborn.common.util.Utils#51 - Successfully started service 'WorkerSys' on port 38303.
```
Closes#1622 from pan3793/CELEBORN-713.
Authored-by: Cheng Pan <chengpan@apache.org>
Signed-off-by: Cheng Pan <chengpan@apache.org>
### What changes were proposed in this pull request?
Change Celeborn Master URL from `rss://<host>:<port>` to `celeborn://<host>:<port>`
### Why are the changes needed?
Respect the project name.
### Does this PR introduce _any_ user-facing change?
Yes, migration guide is updated accordingly.
### How was this patch tested?
Pass GA.
Closes#1624 from pan3793/CELEBORN-715.
Authored-by: Cheng Pan <chengpan@apache.org>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
Remove environment variables `CELEBORN_MASTER_HOST` and `CELEBORN_MASTER_PORT`, and makes `CELEBORN_LOCAL_HOSTNAME` takes effect on both master and worker.
### Why are the changes needed?
There are many different ways to configure the master/worker host and port, which makes the thing complex and inconsistent.
After this change,
#### master
1. cli args `--host` `--port` takes the highest priority
2. then lookup env `CELEBORN_LOCAL_HOSTNAME`
3. things are different when HA enabled and disabled
3.1. when HA is disabled, lookup configurations `celeborn.master.host` and `celeborn.master.port`
3.2. when HA is enabled, each node needs to know the whole cluster info,
```
celeborn.master.ha.node.1.host clb-1
celeborn.master.ha.node.1.port 9097
celeborn.master.ha.node.2.host clb-2
celeborn.master.ha.node.2.port 9097
celeborn.master.ha.node.3.host clb-3
celeborn.master.ha.node.3.port 9097
```
in addition, `celeborn.master.ha.node.id=1` can be used to indicate the node id, otherwise, the master will try to bind each host to match the node id.
#### worker
1. cli args `--host` `--port` takes the highest priority
2. then lookup env `CELEBORN_LOCAL_HOSTNAME`
things are simple than the master case because each worker is not required to know others.
### Does this PR introduce _any_ user-facing change?
Yes.
### How was this patch tested?
UT.
Closes#1616 from pan3793/CELEBORN-707.
Authored-by: Cheng Pan <chengpan@apache.org>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
30s for fetch timeout is too short and easy to exceed. This PR increases the default value to 600s.
### Why are the changes needed?
When I was testing 3T TPCDS with three workers, I encountered fetch timeout:
```
23/06/21 16:46:41,771 INFO [fetch-server-11-7] FetchHandler: Sending chunk 28856864163, 1, 0, 2147483647
...
23/06/21 16:47:16,870 INFO [fetch-server-11-7] FetchHandler: Sent chunk 28856864163, 1, 0, 2147483647
```
And I remember from some users' monitoring, the max fetch time can reach several minutes on heavy load without error.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Manual test.
Closes#1618 from waitinfuture/709.
Authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
Extend doc about migration from 0.2.1 to 0.3.0. Added the following contents:
<img width="1084" alt="image" src="https://github.com/apache/incubator-celeborn/assets/26535726/7a9d172c-09ba-48b6-9f5c-73a8b13d035f">
### Why are the changes needed?
ditto
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
No.
Closes#1612 from waitinfuture/702.
Lead-authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Co-authored-by: Cheng Pan <pan3793@gmail.com>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
It was discussed during the last meeting, but abandoned due to the complication.
### Why are the changes needed?
Make the configuration unified.
### Does this PR introduce _any_ user-facing change?
Yes, but the legacy configurations still take effect.
### How was this patch tested?
New UTs.
Closes#1549 from pan3793/CELEBORN-638.
Authored-by: Cheng Pan <chengpan@apache.org>
Signed-off-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
### What changes were proposed in this pull request?
Refresh celeborn configurations in doc
### Why are the changes needed?
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Closes#1592 from AngersZhuuuu/CELEBORN-680.
Authored-by: Angerszhuuuu <angers.zhu@gmail.com>
Signed-off-by: Angerszhuuuu <angers.zhu@gmail.com>
### What changes were proposed in this pull request?
Celeborn fetch chunk also should support check timeout
#### Test case
```
executor instance 20
SQL:
SELECT count(1) from (select /*+ REPARTITION(100) */ * from spark_auxiliary.t50g) tmp;
--conf spark.celeborn.client.spark.shuffle.writer=sort \
--conf spark.celeborn.client.fetch.excludeWorkerOnFailure.enabled=true \
--conf spark.celeborn.client.push.timeout=10s \
--conf spark.celeborn.client.push.replicate.enabled=true \
--conf spark.celeborn.client.push.revive.maxRetries=10 \
--conf spark.celeborn.client.reserveSlots.maxRetries=10 \
--conf spark.celeborn.client.registerShuffle.maxRetries=3 \
--conf spark.celeborn.client.push.blacklist.enabled=true \
--conf spark.celeborn.client.blacklistSlave.enabled=true \
--conf spark.celeborn.client.fetch.timeout=30s \
--conf spark.celeborn.client.push.data.timeout=30s \
--conf spark.celeborn.client.push.limit.inFlight.timeout=600s \
--conf spark.celeborn.client.push.maxReqsInFlight=32 \
--conf spark.celeborn.client.shuffle.compression.codec=ZSTD \
--conf spark.celeborn.rpc.askTimeout=30s \
--conf spark.celeborn.client.rpc.reserveSlots.askTimeout=30s \
--conf spark.celeborn.client.shuffle.batchHandleChangePartition.enabled=true \
--conf spark.celeborn.client.shuffle.batchHandleCommitPartition.enabled=true \
--conf spark.celeborn.client.shuffle.batchHandleReleasePartition.enabled=true
```
Test with 3 worker and add a `Thread.sleep(100s)` before worker handle `ChunkFetchRequest`
Before patch
<img width="1783" alt="截屏2023-06-14 上午11 20 55" src="https://github.com/apache/incubator-celeborn/assets/46485123/182dff7d-a057-4077-8368-d1552104d206">
After patch
<img width="1792" alt="image" src="https://github.com/apache/incubator-celeborn/assets/46485123/3c8b7933-8ace-426d-8e9f-04e0aabfac8e">
The log shows the fetch timeout checker workers
```
23/06/14 11:14:54 ERROR WorkerPartitionReader: Fetch chunk 0 failed.
org.apache.celeborn.common.exception.CelebornIOException: FETCH_DATA_TIMEOUT
at org.apache.celeborn.common.network.client.TransportResponseHandler.failExpiredFetchRequest(TransportResponseHandler.java:147)
at org.apache.celeborn.common.network.client.TransportResponseHandler.lambda$new$1(TransportResponseHandler.java:103)
at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308)
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180)
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
at java.lang.Thread.run(Thread.java:748)
23/06/14 11:14:54 WARN RssInputStream: Fetch chunk failed 1/6 times for location PartitionLocation[
id-epoch:35-0
host-rpcPort-pushPort-fetchPort-replicatePort:10.169.48.203-9092-9094-9093-9095
mode:MASTER
peer:(host-rpcPort-pushPort-fetchPort-replicatePort:10.169.48.202-9092-9094-9093-9095)
storage hint:StorageInfo{type=HDD, mountPoint='/mnt/ssd/0', finalResult=true, filePath=}
mapIdBitMap:null], change to peer
org.apache.celeborn.common.exception.CelebornIOException: Fetch chunk 0 failed.
at org.apache.celeborn.client.read.WorkerPartitionReader$1.onFailure(WorkerPartitionReader.java:98)
at org.apache.celeborn.common.network.client.TransportResponseHandler.failExpiredFetchRequest(TransportResponseHandler.java:146)
at org.apache.celeborn.common.network.client.TransportResponseHandler.lambda$new$1(TransportResponseHandler.java:103)
at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308)
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180)
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
at java.lang.Thread.run(Thread.java:748)
Caused by: org.apache.celeborn.common.exception.CelebornIOException: FETCH_DATA_TIMEOUT
at org.apache.celeborn.common.network.client.TransportResponseHandler.failExpiredFetchRequest(TransportResponseHandler.java:147)
... 8 more
23/06/14 11:14:54 INFO SortBasedShuffleWriter: Memory used 72.0 MB
```
### Why are the changes needed?
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Closes#1587 from AngersZhuuuu/CELEBORN-676.
Authored-by: Angerszhuuuu <angers.zhu@gmail.com>
Signed-off-by: Angerszhuuuu <angers.zhu@gmail.com>
### What changes were proposed in this pull request?
Add celeborn.metrics.conf to conf entity
### Why are the changes needed?
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Closes#1593 from AngersZhuuuu/CELEBORN-681.
Authored-by: Angerszhuuuu <angers.zhu@gmail.com>
Signed-off-by: Angerszhuuuu <angers.zhu@gmail.com>
### What changes were proposed in this pull request?
refer to https://github.com/apache/spark/pull/40301
1. Optimize `Utils.bytesToString`. Arithmetic ops on BigInt and BigDecimal are order(s) of magnitude slower than the ops on primitive types. Division is an especially slow operation and it is used en masse here.
2. According to the information sourced from [Wikipedia](https://en.wikipedia.org/wiki/Kilobyte), it is established that 1000 is the appropriate factor for representing kilobytes (KB), while 1024 is the correct factor for kibibytes (KiB). In alignment with this understanding, changing the size unit from "KB" to "KiB".
### Why are the changes needed?
the Utils#bytesToString method is frequently employed in memory-related log messages.
### Does this PR introduce _any_ user-facing change?
No, only perf improvement.
### How was this patch tested?
existing UT and manually tested.
Closes#1590 from cfmcgrady/bytesToString.
Authored-by: Fu Chen <cfmcgrady@gmail.com>
Signed-off-by: Cheng Pan <chengpan@apache.org>
### What changes were proposed in this pull request?
Give Heartbeat one byte message and skip this byte when decode.
### Why are the changes needed?
Heartbeat message may split in to two netty buffer, then the `empty buffer` (which don't need actually, but need keep) be wrong removed, then decodeNext would throw NPE. see
``` java
while (headerBuf.readableBytes() < HEADER_SIZE) {
ByteBuf next = buffers.getFirst();
int toRead = Math.min(next.readableBytes(), HEADER_SIZE - headerBuf.readableBytes());
headerBuf.writeBytes(next, toRead);
if (!next.isReadable()) {
buffers.removeFirst().release();
}
}
```
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
UT & MANUAL
Closes#1589 from RexXiong/CELEBORN-675.
Authored-by: Shuang <lvshuang.tb@gmail.com>
Signed-off-by: zhongqiang.czq <zhongqiang.czq@alibaba-inc.com>
### What changes were proposed in this pull request?
CommitHandler will check whether the target worker is in WorkerStatusTracker's excluded list. If so, skip calling commit files on it.
### Why are the changes needed?
Avoid unnecessary commit files to excluded worker.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Manual test.
Closes#1581 from waitinfuture/669.
Lead-authored-by: zky.zhoukeyong <zky.zhoukeyong@alibaba-inc.com>
Co-authored-by: Angerszhuuuu <angers.zhu@gmail.com>
Co-authored-by: Keyong Zhou <zhouky@apache.org>
Signed-off-by: Shuang <lvshuang.tb@gmail.com>
### What changes were proposed in this pull request?
In our prod meet many times of push queue stuck caused by PushState's status was not being removed.
Caused DataPushQueue to keep waiting for taking task.
Although have resolved some bugs, here we'd better add a max wait time for taking tasks since we already have the `PUSH_DATA_TIMEOUT` check method. If the target worker is really stuck, we can retry our task finally.
### Why are the changes needed?
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Closes#1552 from AngersZhuuuu/CELEBORN-640.
Authored-by: Angerszhuuuu <angers.zhu@gmail.com>
Signed-off-by: Angerszhuuuu <angers.zhu@gmail.com>
### What changes were proposed in this pull request?
Pluginconf might be hard to understand why Celeborn needs to config class.
### Why are the changes needed?
Ditto.
### Does this PR introduce _any_ user-facing change?
NO.
### How was this patch tested?
UT.
Closes#1524 from FMX/CELEBORN-610.
Authored-by: Ethan Feng <ethanfeng@apache.org>
Signed-off-by: Ethan Feng <ethanfeng@apache.org>
### What changes were proposed in this pull request?
Add doc about enabling rac-awareness
### Why are the changes needed?
Document new features.
### Does this PR introduce _any_ user-facing change?
Yes, the docs are updated.
### How was this patch tested?
<img width="1085" alt="截屏2023-06-02 下午3 19 10" src="https://github.com/apache/incubator-celeborn/assets/46485123/c8c51a4c-40be-40ea-befd-3c369b9f7600">
Closes#1536 from AngersZhuuuu/CELEBORN-629.
Authored-by: Angerszhuuuu <angers.zhu@gmail.com>
Signed-off-by: Cheng Pan <chengpan@apache.org>
### What changes were proposed in this pull request?
Refine this doc since:
1. It didn't mention our cluster default RPC type is `NETTY`
2. If the user use the ratis shell with `GRPC` but didn't know the ratis cluster is `NETTY`, the error is not clear and hard to debug.
### Why are the changes needed?
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Closes#1542 from AngersZhuuuu/CELEBORN-623-FOLLOWUP.
Lead-authored-by: Angerszhuuuu <angers.zhu@gmail.com>
Co-authored-by: Cheng Pan <pan3793@gmail.com>
Signed-off-by: Cheng Pan <chengpan@apache.org>
