Fixed bug in exponential retry intervals

[britzl]

Fixed crash when creating an exponential retry interval. Also added tests for the retries.lua module.

./nakama/util/retries.lua:12: attempt to perform arithmetic on a nil value (field '?')
stack traceback:
	./nakama/util/retries.lua:12: in function 'nakama.util.retries.exponential'
	test/test_retries.lua:11: in function <test/test_retries.lua:10>
	[C]: in function 'xpcall'
	./telescope.lua:418: in local 'invoke_test'
	./telescope.lua:458: in function 'telescope.run'
	./tsc:274: in main chunk
	[C]: in ?

[britzl]

@DannyIsYog could you please review this PR and the other's in this repo assigned to you? Thanks!

[DannyIsYog]

Hey @britzl this looks good!

I just have a suggestions for a future PR.
I'm not very familiar with retries so I was looking at how it's done on our other sdks and noticed most (if not all) use retries with jitter. This is, if a large number of clients all start retrying a failed service at the same time, this predictable sequence can lead to a problem, where all clients hit the service simultaneously after the same delay, causing it to fail again, since the pattern is "predictable". To fix this, we should add a bit of randomness to this so to "spread" when the clients will re-hit the server.

You can see something like that for the Nakama Unity SDK

Let me know if this makes sense.

[britzl]

Let me know if this makes sense.

It totally does. I added an issue to track this: #86

[britzl]

Seems like leafo/gh-actions-lua has some issues since a few months back. There's a PR to fix it but the author has not merged it. We might as well just install Lua using apt-get.