[
https://issues.apache.org/jira/browse/RATIS-1277?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17256865#comment-17256865
]
runzhiwang commented on RATIS-1277:
-----------------------------------
[~szetszwo] This problem caused by the key of NettyClientStreamRpc#replies
which should not use streamId as the key, it should use clientId + streamId as
key. In my previous test, different client choose different primary, and
primary connect to two peers, so the NettyClientStreamRpc#replies of primary
does not accpet request from different clients, so this problem did not happen
in my previous test. But now, we use primary -> peer1 -> peer2, though
different client choose different primary, but peer1 can accpet request from
different clients, then reply and request are out of order. We can find from
the following message, the reply and request has the same streamId, but has
different clientId.
{code:java}
succ:true reply written:0 expected:1000000
clientId:client-6C3D34F237DA,type:STREAM_HEADER,streamId4,offset:0,datalength:0
localWrite:1927248929 remoteWrites:789787006
request:DataStreamRequestByteBuf:clientId=client-F1F2794DEFF0,type=STREAM_DATA,id=4,offset=48000000,length=1000000
{code}
> FileStore write failed because out of order
> -------------------------------------------
>
> Key: RATIS-1277
> URL: https://issues.apache.org/jira/browse/RATIS-1277
> Project: Ratis
> Issue Type: Sub-task
> Reporter: runzhiwang
> Assignee: runzhiwang
> Priority: Major
> Attachments: screenshot-2.png, screenshot-3.png
>
>
> !screenshot-3.png!
> As the following image and code shows, the code check the byteWritten of
> STREAM_HEADER, i.e. 0, equals to 10000, of course failed.
> !screenshot-2.png!
> {code:java}
> static boolean
> checkSuccessRemoteWrite(List<CompletableFuture<DataStreamReply>>
> replyFutures, long bytesWritten) {
> for (CompletableFuture<DataStreamReply> replyFuture : replyFutures) {
> final DataStreamReply reply = replyFuture.join();
> if (!reply.isSuccess() || reply.getBytesWritten() != bytesWritten) {
> + System.err.println("succ:" + reply.isSuccess() + " reply written:"
> + reply.getBytesWritten() +
> + " expected:" + bytesWritten + " clientId:" + reply.getClientId()
> + ",type:" + reply.getType() + ",streamId" +
> + reply.getStreamId() + ",offset:" + reply.getStreamOffset() +
> ",datalength:" + reply.getDataLength());
> return false;
> }
> }
> return true;
> }
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
