Extending deadline for logging

[r...@google.com]

I'm working on a service that write some log info to spanner when it's done responding to a request. 

However, the service uses the context's deadline to write to spanner, so if the deadline expires due to some long running RPC, the write to spanner fails (because the deadline expired), and we don't get any log info.

What's the best practice for dealing with this situation?

[r...@google.com]

As an example:

Client calls MyService with a deadline of 10 seconds.

MyService calls OtherService as part of responding. However, the call to OtherService times out due to the deadline in 10 seconds.

MyService tries to log the error to Spanner; but it's still using that context deadline which expired.

Is there a way to get a new context with an extended deadline? Are there any issues with that approach?

[Dave Cheney]

On Thursday, 16 August 2018 12:15:57 UTC+10, r...@google.com wrote:

As an example:

Client calls MyService with a deadline of 10 seconds.

MyService calls OtherService as part of responding. However, the call to OtherService times out due to the deadline in 10 seconds.

MyService tries to log the error to Spanner; but it's still using that context deadline which expired.

Is there a way to get a new context with an extended deadline? Are there any issues with that approach?

Sure, just call context.WithDeadline(context.Background()) and use that instead

 The difficulty is you want the deadline of this new context to live beyond its parent. Logically it feels like this new context is subordinate to the previous, but by design we've said that the new context is _not_ subordinate -- the deadline does not apply to it. I think you need to address this incongruousness before proceeding.

[Robert Bartoszynski]

Thanks. Perhaps an alternative would be for me to create a child context with a deadline of (context - x seconds) and pass that to OtherService, with the expectation that there should be x seconds left over for the write to spanner.

--
You received this message because you are subscribed to a topic in the Google Groups "golang-nuts" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/golang-nuts/TUicHyvYNX0/unsubscribe.
To unsubscribe from this group and all its topics, send an email to golang-nuts...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Dave Cheney]

What would happen if you write the error to spanner with a setting
context.Backgrond(), ie, no deadline?

[Dave Cheney]

Thinking some more about the problem, I think your solution of reserving some of the deadline to handle the spanner log in the error case sounds like the best answer. However it does lead to questions like, if you reserve n seconds to log to spanner in the error path, and it takes longer than n, what happens to the error, is it dropped?

>> golang-nuts+unsubscribe@googlegroups.com.

[Robert Bartoszynski]

Not sure what would happen if I set the context to context.Background(), but I'd have to understand the implications of it really well before I did something against the grain like that (using context.Background() for request-specific code).

Yeah, perhaps we'll go with the solution of reserving time for Spanner. I figure I can find out the different percentiles for how long the Spanner writes take and then set the amount of time reserved for Spanner accordingly. It will probably be impossible to guarantee that logs never get dropped, but we can probably make them exceedingly rare.

>> golang-nuts...@googlegroups.com.

>> For more options, visit https://groups.google.com/d/optout.

--

You received this message because you are subscribed to a topic in the Google Groups "golang-nuts" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/golang-nuts/TUicHyvYNX0/unsubscribe.

To unsubscribe from this group and all its topics, send an email to golang-nuts...@googlegroups.com.