net/http: panic: runtime error: slice bounds out of range #12339
Labels
Milestone
Comments
|
Thank you for this bug report. Have you confirmed your code is free of data races ? If so, can you please include some sample code that demonstrates the issue so that others can try to confirm the issue you are seeing. |
|
Code used to replicate issue seen in another application (with URLs redacted): -race yields: EDIT: changed the code above to only access resp if RoundTrip was successful. Reran test to confirm that the issue persists (it does). |
|
Thank you for providing your sample code. Although it should not matter for this issue, I cannot see where you are closing the response body. /cc @bradfitz |
|
I wanted a small piece of example code exhibiting the problem I saw in another application. I should have taken more care in writing it. I've now tested the example above with resp.Body.Close(). The problem persists. |
|
I'm not clear on your initial report: does this failure happen with 1.4.2, or only with 1.5? |
|
It happens on both 1.4.2 and 1.5. EDIT: paths above relating to /home/sebastian/go is go 1.5 |
|
Can you email me privately the URL(s) I can hit to reproduce this? I am |
|
Email sent |
|
I can't reproduce. Can you reproduce with Go tip? (what is to become Go 1.6 soon-ish) |
|
With same code/line-numbers as in email, using master (assuming master == tip): |
|
@sebcat, can you try again with the latest code from master? I was never able to reproduce this myself but I wonder if one my recent commits might have fixed this. The same output as before would be great. Thanks! |
|
(In particular, I think https://golang.org/cl/17890 could've fixed this) |
|
I am afraid that I am still able to reproduce. I'm looking into the possibility to set up a reproducible test environment. I tried reproducing it on a virtual instance at DO, but no luck (maybe related to difference in latency/different responses from server due to geoloc? purely speculation). |
|
Thanks for the update. The updated crash with new line numbers helps. I'll look again. |
|
Sorry, I haven't forgotten about this. I'm just getting back to things after the holidays. Can I trouble you to try once again from master? Some changes just went in which have a small but non-zero chance of fixing this. If you can reproduce the race again (or not), please include the git version number in your reply. Thanks! |
|
I am no longer able to reproduce the problem. Building with -race and executing produces no output from the race detector and no panic occurs over multiple runs. Thank you! |
|
Excellent. Thank you for your help testing. For the record, I suspect the change which fixed it was git rev 7fa9846 (https://go-review.googlesource.com/18282) or less likely 7de71c8 (https://go-review.googlesource.com/18260) |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Hello,
I experience a panic intermittently when using a Transport concurrently against certain URLs. It happens when I read the response body using an io.LimitedReader with ioutil.ReadAll.
I've isolated the problem from a larger application and I would like to share the code for issue reproduction privately, as I am only able to reproduce it for certain URLs and I don't want those URLs in the issue tracker.
I've tested the example with go 1.4.2 and go 1.5 on a couple of different machines. If I run the example locally, I may have to run it a few times before experiencing a panic. If I run the example on a remote machine with different network/hardware/linux distribution that I have access to, it occurs 100% of the time. I don't know why.
Complete stack trace:
The text was updated successfully, but these errors were encountered: