I just thought I'd close this discussion up a bit as the PR is now merged.
I think it's worth noting that as long as the TinkerPop Transaction API
does not have a universally accepted method for passing around transaction
state among threads we will always have a situation where we need to keep a
s
I've opened a PR for the UnifiedChannelizer:
https://github.com/apache/tinkerpop/pull/1414
Over the weekend I passed 10s of millions of messages through the
UnifiedChannelizer and the server stayed quite stable throughout it all.
With all the tests passing nicely now in travis, docker and locally
I've solved the Travis issue. Some tests simply required more threads in
gremlinPool than the default which is set to the available cores. On Travis
you probably get a machine with 2 cores available and some tests required 3
or more to keep that many sessions open at once. Silly problem though the
I've been fussing with Travis for a while now haven't had much luck
figuring out what is wrong. Decided to start profiling the server to see if
that provided any hints (and it was something I intended to do in any
case). Two issues came up:
1. shouldEventuallySucceedOnSameServerWithDefault() had s
I can't seem to get Travis to pass the build for this branch with the
UnifiedChannelizer enabled.
https://travis-ci.com/github/apache/tinkerpop/builds/221427334
It just hangs without much information. I can't seem to get it to behave
otherwise and it doesn't fail for me locally (of course). If an
It took me a long time to get to the final steps of this change. Uncovered
some unexpected bugs in the WsAndHttpChannelizer and that was gnarly to
sort out. Also found some problems with authorization through the
SaslAndHttpBasicAuthenticationHandler. I can't say I really like how these
combined ch
More reasonable progress with the UnifiedChannelizer. The short version is
that I have a pretty major quantity of the server integration tests passing
under this new Channelizer. The ones that are failing are typically ones
that assert some specific log messages or exception message which may not
n
I've started to play with some ideas for pushing all requests
scripts/bytecode sessions/sessionless down one unified pipeline with one
shared thread pool. The most interesting thing I've seen so far is how much
the code could potentially simplify under this approach. The
GremlinExecutor was the wro
>
> I suppose the
> question is how would we ensure that each request for a session ends up
> being executed on the same thread from the previous request if jobs are
> randomly submitted to a worker pool?
I haven't thought through the details, but on top of my head, we would have
to maintain some
I started a fresh thread for this topic Divij brought up, with more context:
> In a scenario where we have both
> session-less and sesion-ful requests being made to the server, the
resource
> allocation will not be fair and may lead to starvation for some requests.
> This problem exists even today
10 matches
Mail list logo