Github user tnachen commented on a diff in the pull request:
https://github.com/apache/spark/pull/11272#discussion_r54507553
--- Diff:
network/shuffle/src/main/java/org/apache/spark/network/shuffle/mesos/MesosExternalShuffleClient.java
---
@@ -53,21 +65,58 @@ public MesosExternalShuffleClient(
super(conf, secretKeyHolder, saslEnabled, saslEncryptionEnabled);
}
- public void registerDriverWithShuffleService(String host, int port)
throws IOException {
+ public void registerDriverWithShuffleService(
+ String host,
+ int port,
+ long heartbeatTimeoutMs,
+ long heartbeatIntervalMs) throws IOException {
+
checkInit();
- ByteBuffer registerDriver = new RegisterDriver(appId).toByteBuffer();
+ ByteBuffer registerDriver = new RegisterDriver(appId,
heartbeatTimeoutMs).toByteBuffer();
TransportClient client = clientFactory.createClient(host, port);
- client.sendRpc(registerDriver, new RpcResponseCallback() {
- @Override
- public void onSuccess(ByteBuffer response) {
- logger.info("Successfully registered app " + appId + " with
external shuffle service.");
- }
-
- @Override
- public void onFailure(Throwable e) {
- logger.warn("Unable to register app " + appId + " with external
shuffle service. " +
+ client.sendRpc(registerDriver, new RegisterDriverCallback(client,
heartbeatIntervalMs));
+ }
+
+ private class RegisterDriverCallback implements RpcResponseCallback {
+ private final TransportClient client;
+ private final long heartbeatIntervalMs;
+
+ private RegisterDriverCallback(TransportClient client, long
heartbeatIntervalMs) {
+ this.client = client;
+ this.heartbeatIntervalMs = heartbeatIntervalMs;
+ }
+
+ @Override
+ public void onSuccess(ByteBuffer response) {
+ heartbeaterThread.scheduleAtFixedRate(
+ new Heartbeater(client), 0, heartbeatIntervalMs,
TimeUnit.MILLISECONDS);
+ logger.info("Successfully registered app " + appId + " with external
shuffle service.");
+ }
+
+ @Override
+ public void onFailure(Throwable e) {
+ logger.warn("Unable to register app " + appId + " with external
shuffle service. " +
"Please manually remove shuffle data after driver exit. Error: "
+ e);
- }
- });
+ }
+ }
+
+ @Override
+ public void close() {
+ heartbeaterThread.shutdownNow();
+ super.close();
+ }
+
+ private class Heartbeater implements Runnable {
+
+ private final TransportClient client;
+
+ private Heartbeater(TransportClient client) {
+ this.client = client;
+ }
+
+ @Override
+ public void run() {
+ client.send(new ShuffleServiceHeartbeat(appId).toByteBuffer());
--- End diff --
In the case where network problems occurred and the Shuffle service times
out and already drops the app, if the driver is still running and heart beats
eventually land, perhaps as an optimization we can also tell the Heartbeater to
stop heartbeating since the app is not recognized anymore. Not necessary in
this patch but I'm thinking might be worth leaving a TODO.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
