Lazy behaviour when parallel requests

Question

Lazy behaviour when parallel requests

pirhoo opened this issue 11 years ago · comments

Pierre Romera Zhang commented 11 years ago

Hi,

I have a curious behaviour that I still can't explain after 2 days of investigation. When I execute several queries with neo4django at the same time (or very close), the process gets lazy and the request never ends (probably because the previous one is not done).

I got this bug with neo4j 1.8.2 and 1.9.1, and Django 1.4.4 and 1.5.4. I can't give more detail because the only way to provoke this bug is to madly refresh a page.

Thanks for the amazing work,
Pierre

Matt Luongo · Answer 1 · Sat Oct 26 2013 05:18:38 GMT+0800 (China Standard Time)

Hm, there is some transactional logic related to the in-graph type tree- maybe that's involved? What sort of requests are they, read or write?

Pierre Romera Zhang · Answer 2 · Sat Oct 26 2013 05:32:33 GMT+0800 (China Standard Time)

Mainly read and without any particular conditions: I succeed to reproduce this bug with a simple User.objects.all() (among 20 nodes).

Matt Luongo · Answer 3 · Sat Oct 26 2013 05:37:05 GMT+0800 (China Standard Time)

Hm, that's no good. Is it using the dev server or in production? If it were reproducible using, say, the requests library and threads, that'd be easier to test & fix.

Pierre Romera Zhang · Answer 4 · Sat Oct 26 2013 05:56:12 GMT+0800 (China Standard Time)

Interesting: it seems that the bug doesn't happen in production (on heroku).

Matt Luongo · Answer 5 · Sat Oct 26 2013 06:03:20 GMT+0800 (China Standard Time)

Aha! Good, I was worried. It might be because the Django dev server is
threaded, while production is usually process-based. To test that I believe
there's a '--nothreading' (or similarly named) argument to runserver.
Maybe give that a shot? Regardless this sounds like a bug, but we'll see
whether it's a big or little one.
On Oct 25, 2013 5:56 PM, "Pierre Romera" notifications@github.com wrote:

Interesting: it seems that the bug doesn't happen in production (on
heroku).

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/223#issuecomment-27129143
.

Pierre Romera Zhang · Answer 6 · Sat Oct 26 2013 06:07:39 GMT+0800 (China Standard Time)

OK, that sounds reassuring. I'm trying what you propose.

tonjo · Answer 7 · Sat Oct 26 2013 06:13:46 GMT+0800 (China Standard Time)

Hmm, and threading would worsen performance?
Il 26/ott/2013 00:07 "Pierre Romera" notifications@github.com ha scritto:

OK, that sounds reassuring. I'm trying what you propose.

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/223#issuecomment-27129742
.

Pierre Romera Zhang · Answer 8 · Sat Oct 26 2013 06:19:44 GMT+0800 (China Standard Time)

Ha! Good news: the --nothreading option fix it, thanks!

I don't think we can consider this issue as solved so I let you decide if we should close it.

Matt Luongo · Answer 9 · Sat Oct 26 2013 06:21:26 GMT+0800 (China Standard Time)

No, I'll leave it open. It's going to be fun to debug :)
On Oct 25, 2013 6:19 PM, "Pierre Romera" notifications@github.com wrote:

Ha! Good news: the --nothreadinghttps://docs.djangoproject.com/en/dev/ref/django-admin/#django-admin-option---nothreadingoption fix it, thanks!

I don't think we can consider this issue as solved so I let you decide if
we should close it.

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/223#issuecomment-27130367
.

Pierre Romera Zhang · Answer 10 · Sat Oct 26 2013 06:23:11 GMT+0800 (China Standard Time)

OK, you're the boss. Thanks again, you saved my weekend!