-
Notifications
You must be signed in to change notification settings - Fork 1.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
BrokenPipeErrors while using datastore (dev) #1811
Comments
Thanks for reporting @jennolsen84! @dhermes, is this related to the issues we were talking about in #1796? |
@jennolsen84 This seems to be #1214. Any chance your application is threaded? @jonparrott and I have been working to move off of |
@dhermes I think we're seeing this in single threaded apps as well as multithreaded ones. Probably not too hard to reproduce... I assume you already have a dev datastore setup, so you just need to connect to it (ideally on a different machine so you can tcpdump), do a query, and then wait 5 minutes, and then try to do another query, and you should see the exception. |
This is just part of the nature of long connections. We should retry requests on connections errors. A similar issue was brought up in google-api-python-client. |
@jonparrott Is this an issue that |
I actually don't know if urllib3/requests does that automatically. |
@jennolsen84 @jonparrott I'm not sure what the "right" thing to do is for users. So far, we've taken the stance that service errors (e.g. a 500, 502, 503 from Google) are errors the user should be able to see. This one somewhat falls into that category: the connection timed out, all the time we waited for it was wasted. Maybe we could wrap it and provide a better exception? (We certainly need a retry story and this would fit into it, along with 429 errors, 404s from eventual consistency failures, etc.) @tseaver WDYT? |
@dhermes perhaps a Another example to study is: But, I have never really felt great about the way that SQL Alchemy handles it. |
I'm still not sure what the right thing to do is here, since a broken pipe is a few layers below our library. @tseaver WDYT? |
I think wrapping it in an exception defined in |
The problem is "it". Wrapping "it" is ill-defined if people are bringing their own transport layer. In some sense we can catch something caused by the low-level |
This SO answer goes over how retries are usually done with requests/urllib3, perhaps it helps: http://stackoverflow.com/questions/21371809/cleanly-setting-max-retries-on-python-requests-get-or-post-method I think it would be great to get some of this functionality built in to gcloud-python, and then easily be able to disable the retry mechanism if desired. This will help users write less retry logic in their app, and can focus on just error handling when service is truly unavailable, etc. |
This seems like more fuel to the fire of ditching httplib2. On Sat, May 28, 2016, 11:25 PM jennolsen84 notifications@github.com wrote:
|
Totally. We should have a pow-wow and get |
Hello, As part of trying to get things under control (as well as to empower us to provide better customer service in the future), I am declaring a "bankruptcy" of sorts on many of the old issues, especially those likely to have been addressed or made obsolete by more recent updates. My goal is to close stale issues whose relevance or solution is no longer immediately evident, and which appear to be of lower importance. I believe in good faith that this is one of those issues, but I am scanning quickly and may occasionally be wrong. If this is an issue of high importance, please comment here and we will reconsider. If this is an issue whose solution is trivial, please consider providing a pull request. Thank you! |
We've recently started to see more BrokenPipeErrors while using gcloud-python. We started noticing them more when we upgraded to Kubernetes 1.2.4 (vs 1.1), if that makes any difference. But, I think this was just a coincidence.
Anyway, I delved deeper and saw that the dev datastore version was sending FIN ACK, and then RST ACK packets after some inactivity. The datastore client machine sees these packets as well. The next time the client tries to do a query against datastore, we get a broken pipe error. It would be nice if the gcloud-python / httplib2 did some pre-emptive checking to see if the connection was closed retried the query itself.
Some code snippets:
This is one way to detect if the socket on the other side has closed a connection.
Here is a stack trace of the exception
The text was updated successfully, but these errors were encountered: