support for max_time as an alternative to max_tries about backoff

[bgreen-litl]

@terrisgit @nirtayeb @cjw296 max_time is now supported in master - I will probably release as backoff=1.5

[bgreen-litl]

@tjboring thanks for the nudge. backoff==1.5.0 is now released and in pypi

Thanks all.

[bgreen-litl]

@terrisgit You can use max_tries to effectively set this maximum (though it might be inconvenient because you need to do the math taking into account the wait generator you're using) but @cjw296 is correct that when using full jitter max_tries is not deterministic in defining the minimum wait time.

For example if you specify max_tries=3 and get three (un?)lucky full jitter rolls, you might retry for a significantly shorter overall period than expected based on your wait generator.

I'll plan to add a max_time keyword arg.

[terrisgit]

Yes, a maximum is needed to avoid crazy waits like 10,000 seconds

[bgreen-litl]

@cjw296 There's a max_time implementation here #48

I'll give it some more testing before I merge is to master, but let me know if you have any feedback in the meantime.

[terrisgit]

Awesome! I'll give it a try.

max_tries in my use case is infinity, or until the process receives SIGINT. I guess I should submit another issue for infinity. I'm just using 1 billion for now.

[cjw296]

@bgreen-litl - thanks! Unfortunately, it's been so long since I submitted this issue that I no longer have the use case to hand. Looks like others are able to test, though, so good luck!

[bgreen-litl]

@terrisgit The default is infinite retries - simply don't specify max_tries.

BTW, if you're concerned with the individual wait times (as opposed to total retry time) max_value is already supported.

Also, sounds like the SIGINT thing is a problem for async - in my experience everything shuts down cleanly in the synchronous case. I need to put some thought into and do some research on the best way to handle shutdown for async.

[terrisgit]

Me want max_tries=3 when I unit test, infinity when I do not. I never thought of trying None.

[terrisgit]

The SIGINT thing is a problem when the sleep timeout is large. Due to my ignorance of max_value, this was a bigger deal for me than it is now. The request to stop sleep for coroutines only is #47 (if you don't already have your own signal handler, you probably don't care about this issue)

[nirtayeb]

I would benefit from having it as well

[terrisgit]

I will be trying the max_time PR soon.

any sense of when the release will be made? just curious, as i'm looking forward to using it.

👍