-
Notifications
You must be signed in to change notification settings - Fork 9.2k
HADOOP-19295. S3A: fs.s3a.connection.request.timeout too low: quick fix #7087
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
HADOOP-19295. S3A: fs.s3a.connection.request.timeout too low: quick fix #7087
Conversation
...for large uploads over slow links * New default value of 15 minutes in source * Updated docs * Troubleshooting: new stack trace and details of the problem Change-Id: If3319268a8df35b5f0377f31ea9df2c2f49186ea
|
|
||
| /** | ||
| * Default duration of a request before it is timed out: 60s. | ||
| * Default duration of a request before it is timed out: 16m. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
16m seems to mismatch with the actual code, 15 minutes, and performance.md. Maybe, 15m?
| * (default 15s) allowed for a fast failure of all other operations. | ||
| * In V2 SDK this now applies to all operations, including uploads. | ||
| * A large timeout is now needed, even though it means that some service/network | ||
| * failures may now take a long time to surface. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
However, 15minutes sounds like too big as a timeout, doesn't it? What is the background of this huge value?
Technically, 15m means 1 hour when we retry 3 times.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
problem is we want a duration which is valid with many other streams competing for that bandwidth, so a single block upload of, say, 64M needs time. I don't know what a good time is.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
we could maybe cut to something a bit lower, like say 10m?
side issue: why don't you try it as is for a remote upload of the rc1 tar.gz and see how you get on without this patch.
|
💔 -1 overall
This message was automatically generated. |
|
In #7089 I'm doing a real fix with different timings but
for this reason, I think we should target a 3.4.2 with it |
|
|
||
| <!--- global properties --> | ||
|
|
||
| <property> |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
not sure how i deleted this; will revert
steveloughran
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
going to switch to 10m if that is enough for the problem to go away locally
...for large uploads over slow links
I don't think this should be the final patch.
Because I've just discovered how we can set different timeouts on each request, so
a long timeout for data put/post seems straightforward, as well as testing.
How was this patch tested?
big bulk uploads
For code changes:
LICENSE,LICENSE-binary,NOTICE-binaryfiles?