-
Notifications
You must be signed in to change notification settings - Fork 153
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[#1538] feat(spark): report blockIds to spark driver optionally #1677
Conversation
import org.apache.uniffle.common.util.RssUtils; | ||
|
||
/** The class is to manage the shuffle data blockIds in spark driver side. */ | ||
public class BlockIdManager { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I have a little concern about performance. We ever stored all block ids in one shuffle server. It has poor performance. Because bitmap has poor performance.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yes, I will try this and apply this into production to check the performance
@@ -37,6 +37,12 @@ | |||
|
|||
public class RssSparkConfig { | |||
|
|||
public static final ConfigOption<Boolean> RSS_CLIENT_BLOCK_ID_SELF_MANAGED_ENABLED = | |||
ConfigOptions.key("rss.client.blockId.selfManagedEnabled") |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
selfManagementEnabled
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
selfmanaged is adjective. Here the noun is more proper. 👍
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe we should remove client
word to align other blockId
config options.
Except, I'm a little worried about the config options about Uniffle. Because there are some less elegant issues. client
seems extra. Don't use camel style. Proper segments.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe we should remove client word to align other blockId config options.
Done
Except, I'm a little worried about the config options about Uniffle. Because there are some less elegant issues. client seems extra. Don't use camel style. Proper segments.
This could be optimized because we have the discard mechanism in configOption, that need to be detailed in doc.
...test/spark3/src/test/java/org/apache/uniffle/test/ContinuousSelectPartitionStrategyTest.java
Outdated
Show resolved
Hide resolved
client-spark/spark2/src/main/java/org/apache/spark/shuffle/RssShuffleManager.java
Show resolved
Hide resolved
@@ -244,7 +244,7 @@ public int compare(String o1, String o2) { | |||
.get(RSS_CLIENT_BLOCK_ID_SELF_MANAGED_ENABLED); | |||
if (ClientType.GRPC == clientType && !blockIdSelfManagedEnabled) { | |||
// TODO skip validating for GRPC_NETTY, needs to mock ShuffleServerNettyHandler | |||
// TODO: skip validating when blockId is managed in spark driver side. | |||
// skip validating when blockId is managed in spark driver side. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe you can remove the TODO
in the 246 line, too.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
No. todo in 246 is reasonable.
You need to add a document for this. |
Yes I will if you have other comments. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM, wait for your performance report.
This should wait some time 😁 |
Minor update. PTAL @jerqi |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Wait for the performance report.
I think I will merge this. When this feature is appied in our internal cluster, the performance report will be attached. |
Ok for me, too. |
What changes were proposed in this pull request?
Support report blockIds from shuffle-servers to spark driver optionally
Why are the changes needed?
Fix: #1538
Does this PR introduce any user-facing change?
Yes.
rss.client.blockId.selfManagedEnabled
is introduced, default value is false.How was this patch tested?
Integration tests.