Issues in master after libevent merge

sdaftuar commented at 9:35 pm on September 9, 2015: member

Not sure I completely understand this yet, so opening an issue for now:

After #5677, I believe it is now the case that the rpc tests may fail if there’s inactivity for more than 30 seconds after opening a connection. I’m getting repeatable failures locally on some of the longer-running tests (such as bipdersig.py), which I seem to be able to fix by either increasing the rpctimeout on all the nodes in the test, or by keeping the rpc connections warm with more rpc calls.

I think some travis jobs may be failing for this reason as well: https://travis-ci.org/bitcoin/bitcoin/builds/79514930

I’m not sure what the right solution is – should the rpc test framework change somehow to allow for the connection to be re-established? Or should we just be defaulting the timeout to a longer value in the tests that need it?

sdaftuar renamed this:
~~RPC tests failing in master due to rpctimeout?~~
Issues in master after libevent merge
on Sep 14, 2015

morcos commented at 10:31 pm on September 14, 2015: member

Additional issues:

Sometimes this error is printed to stderr on startup, node appears to still work: [warn] getaddrinfo: address family for nodename not supported
rpcbind_test.py does not work

jonasschnelli commented at 10:33 pm on September 14, 2015: contributor

SIGPIPE (when reading a closed socket) needs to be handled as well

ptschip commented at 10:37 pm on September 14, 2015: contributor

I’m seeing also a slightly different message on Native Windows when nodes startup but the node seems to run fine.

[warn] getaddrinfo: no address associated with nodename

On 14/09/2015 3:32 PM, Alex Morcos wrote:

Additional issues:
0Sometimes this error is printed to stderr on startup, node appears
1to still work:
2[warn] getaddrinfo: address family for nodename not supported
0rpcbind_test.py does not work
— Reply to this email directly or view it on GitHub #6655 (comment).

laanwj commented at 3:12 pm on September 21, 2015: member

I noticed that rpcbind.py works with master libevent. So I bisected the rpcbind.py hang to libevent commit https://github.com/libevent/libevent/commit/71e709c7829275a594f767b27468b1b52e8b5bb9

This is a bug in libevent which will be fixed in a next release, thus not something that can be fixed on our side.

dcousens commented at 1:48 am on September 22, 2015: contributor

On another note, is this symptomatic of just not enough testing being done for such a large change?

laanwj commented at 8:06 pm on September 22, 2015: member

@dcousens Using master implies that there can be temporary risks from merges. If you can’t tolerate that, use one of the stable relases (and nothing was really very bad here).

laanwj added the label P2P on Sep 22, 2015

jgarzik commented at 10:54 pm on September 22, 2015: contributor

Sometimes you have to merge something for it to get wide testing and review. “release early, release often”

Better to iterate quickly - merge quickly, get code out to the biggest test lab in the world, The Internet. We can always revert or fix just as quickly.

laanwj commented at 9:57 am on October 7, 2015: member

I’m trying to get the libevent http ipv6 fix backported to 2.2: https://github.com/libevent/libevent/pull/287 In the meantime it would make sense to disable the IPv6 bind test, at least when libevent 2.2 is detected.

laanwj commented at 1:40 pm on October 20, 2015: member

Closing this. Upstream libevent has merged the fix into their stable branch. Until that version is released, the test has been commented out (#6837).

laanwj closed this on Oct 20, 2015

DrahtBot locked this on Sep 8, 2021

Issues in master after libevent merge #6655