[
https://issues.apache.org/jira/browse/CASSANDRA-14358?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16424369#comment-16424369
]
Joseph Lynch commented on CASSANDRA-14358:
------------------------------------------
[~djoshi3] We send RSTs on listen overflows via the {{tcp_abort_on_overflow}}
sysctl, so I don't think that this is a simple listen overflow. You're
absolutely right that nodes coming up can get overwhelmed in 2.1/3.0/3.11, and
I think it's really great that trunk fixes that; but I don't think this is the
bug I've identified here. In this case a Cassandra node that _didn't_ restart
continues trying to send on a dead connection forever until the kernel steps in
and cuts off the connection.
[~jasobrown]
{quote}For trunk, we already use IdleStateHandler in the outbound pipeline for
the "not making progress sending data" case. Take a look at
MessageOutHandler#userEventTriggered. iirc, WriteTimeoutHandler /
ReadTimeoutHandler does not fit our use case for messaging; I'd have to look
again to page it back in.
{quote}
I am definitely not a netty expert, but I think that {{IdleStateHandler}} needs
to be paired with a {{WriteTimeoutHandler}} or {{ReadTimeoutHandler}} in order
to effectively identify such blackhole partitions. The app is sending data on
the socket, it's just never going anywhere. My understanding is that
{{IdleStateHandler}} just sends data if there is none.
{quote}Further, Joseph Lynch, I thought your problem was in pre-4.0? There is a
problem (or, at least, non-optimization) with MessagingService.SocketThread
where the single-threaded accept thread handles receiving the first message pf
the c* internode messaging protocol, which may require executing the TLS
handshake before that can be received (if using internode TLS). Thus a single
new establishing connection may block the accept thread for 500-1000 ms for a
cross-datacenter connection. Even worse, if you are using the
Ec2MultiregionSnitch, all connections for the local region will be torn down
and built anew as we switch from the public IP to the region-local private IP -
more traffic on the accept thread. So one slow connection interaction screws
the whole accept thread.
{quote}
Yes, our problem is with pre-4.0, but I'm 99% sure we can mitigate this by
setting {{tcp_retries2}} to 3 or 5 to tell the OS to cover over the app bug. If
Cassandra trunk is resilient to the (rare) bug I think the sysctl is a
reasonable workaround for earlier versions, what do you think? It's also worth
noting that this kind of partition probably only occurs frequently in stateful
networks that swallow RSTs to unknown flows such as AWS VPC (although as
[~aweisberg] points out, power failure could manifest similarly).
{quote}I have an internal patch for 3.0 which, when a new socket is accepted,
moves all processing (internode protocol handshake, TLS handshake) to a
separate thread (via an executor group). So, if one connection runs into a
problem, the other incoming connections are not held up. I've been reticent
about OSS'ing this late in the 3.0/3.11 branches, even though it's not that
clever or invasive. Let me know if you'd be interested in giving it shot and I
can post it. If it's useful/helpful we can consider committing it, as well.
{quote}
I don't think that would fix this particular bug, but it is awesome and fixes
other issues. I volunteer to test it and review it if that helps get it merged
to 3.0/3.11.
> OutboundTcpConnection can hang for many minutes when nodes restart
> ------------------------------------------------------------------
>
> Key: CASSANDRA-14358
> URL: https://issues.apache.org/jira/browse/CASSANDRA-14358
> Project: Cassandra
> Issue Type: Bug
> Components: Streaming and Messaging
> Environment: Cassandra 2.1.19 (also reproduced on 3.0.15), running
> with {{internode_encryption: all}} and the EC2 multi region snitch on Linux
> 4.13 within the same AWS region. Smallest cluster I've seen the problem on is
> 12 nodes, reproduces more reliably on 40+ and 300 node clusters consistently
> reproduce on at least one node in the cluster.
> So all the connections are SSL and we're connecting on the internal ip
> addresses (not the public endpoint ones).
> Potentially relevant sysctls:
> {noformat}
> /proc/sys/net/ipv4/tcp_syn_retries = 2
> /proc/sys/net/ipv4/tcp_synack_retries = 5
> /proc/sys/net/ipv4/tcp_keepalive_time = 7200
> /proc/sys/net/ipv4/tcp_keepalive_probes = 9
> /proc/sys/net/ipv4/tcp_keepalive_intvl = 75
> /proc/sys/net/ipv4/tcp_retries2 = 15
> {noformat}
> Reporter: Joseph Lynch
> Priority: Major
> Attachments: 10 Minute Partition.pdf
>
>
> I've been trying to debug nodes not being able to see each other during
> longer (~5 minute+) Cassandra restarts in 3.0.x and 2.1.x which can
> contribute to {{UnavailableExceptions}} during rolling restarts of 3.0.x and
> 2.1.x clusters for us. I think I finally have a lead. It appears that prior
> to trunk (with the awesome Netty refactor) we do not set socket connect
> timeouts on SSL connections (in 2.1.x, 3.0.x, or 3.11.x) nor do we set
> {{SO_TIMEOUT}} as far as I can tell on outbound connections either. I believe
> that this means that we could potentially block forever on {{connect}} or
> {{recv}} syscalls, and we could block forever on the SSL Handshake as well. I
> think that the OS will protect us somewhat (and that may be what's causing
> the eventual timeout) but I think that given the right network conditions our
> {{OutboundTCPConnection}} threads can just be stuck never making any progress
> until the OS intervenes.
> I have attached some logs of such a network partition during a rolling
> restart where an old node in the cluster has a completely foobarred
> {{OutboundTcpConnection}} for ~10 minutes before finally getting a
> {{java.net.SocketException: Connection timed out (Write failed)}} and
> immediately successfully reconnecting. I conclude that the old node is the
> problem because the new node (the one that restarted) is sending ECHOs to the
> old node, and the old node is sending ECHOs and REQUEST_RESPONSES to the new
> node's ECHOs, but the new node is never getting the ECHO's. This appears, to
> me, to indicate that the old node's {{OutboundTcpConnection}} thread is just
> stuck and can't make any forward progress. By the time we could notice this
> and slap TRACE logging on, the only thing we see is ~10 minutes later a
> {{SocketException}} inside {{writeConnected}}'s flush and an immediate
> recovery. It is interesting to me that the exception happens in
> {{writeConnected}} and it's a _connection timeout_ (and since we see {{Write
> failure}} I believe that this can't be a connection reset), because my
> understanding is that we should have a fully handshaked SSL connection at
> that point in the code.
> Current theory:
> # "New" node restarts, "Old" node calls
> [newSocket|https://github.com/apache/cassandra/blob/6f30677b28dcbf82bcd0a291f3294ddf87dafaac/src/java/org/apache/cassandra/net/OutboundTcpConnection.java#L433]
> # Old node starts [creating a
> new|https://github.com/apache/cassandra/blob/6f30677b28dcbf82bcd0a291f3294ddf87dafaac/src/java/org/apache/cassandra/net/OutboundTcpConnectionPool.java#L141]
> SSL socket
> # SSLSocket calls
> [createSocket|https://github.com/apache/cassandra/blob/cassandra-3.11/src/java/org/apache/cassandra/security/SSLFactory.java#L98],
> which conveniently calls connect with a default timeout of "forever". We
> could hang here forever until the OS kills us.
> # If we continue, we get to
> [writeConnected|https://github.com/apache/cassandra/blob/6f30677b28dcbf82bcd0a291f3294ddf87dafaac/src/java/org/apache/cassandra/net/OutboundTcpConnection.java#L263]
> which eventually calls
> [flush|https://github.com/apache/cassandra/blob/6f30677b28dcbf82bcd0a291f3294ddf87dafaac/src/java/org/apache/cassandra/net/OutboundTcpConnection.java#L341]
> on the output stream and also can hang forever. I think the probability is
> especially high when a node is restarting and is overwhelmed with SSL
> handshakes and such.
> I don't fully understand the attached traceback as it appears we are getting
> a {{Connection Timeout}} from a {{send}} failure (my understanding is you can
> only get a connection timeout prior to a send), but I think it's reasonable
> that we have a timeout configuration issue. I'd like to try to make Cassandra
> robust to networking issues like this via maybe:
> # Change the {{SSLSocket}} {{getSocket}} methods to provide connection
> timeouts of 2s (equivalent to trunk's
> [timeout|https://github.com/apache/cassandra/blob/11496039fb18bb45407246602e31740c56d28157/src/java/org/apache/cassandra/net/async/NettyFactory.java#L329])
> # Appropriately set recv timeouts via {{SO_TIMEOUT}}, maybe something like 2
> minutes (in old versions via
> [setSoTimeout|https://docs.oracle.com/javase/8/docs/api/java/net/Socket.html#setSoTimeout-int-],
> in trunk via
> [SO_TIMEOUT|http://netty.io/4.0/api/io/netty/channel/ChannelOption.html#SO_TIMEOUT]
> # Since we can't set send timeouts afaik (thanks java) maybe we can have
> some kind of watchdog that ensures OutboundTcpConnection is making progress
> in its queue and if it doesn't make any progress for ~30s-1m, forces a
> disconnect.
> If anyone has insight or suggestions, I'd be grateful. I am going to rule out
> if this is keepalive duration by setting tcp_keepalive_probes to like 1 and
> maybe tcp_retries2 to like 8 get more information about the state of the tcp
> connections the next time this happens. It's a very rare bug and when it does
> happen I only have 10 minutes to jump on the nodes and fix it before it fixes
> itself so I'll do my best.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
