Dynamic trailing computation

[marco6]

This PR adds a non-default strategy to dynamically compute the amount of entries to keep in memory based on the size of the snapshot.

The problem with using a fixed number is that each entry has a different size (depending on the number of pages written for each transaction). As such, the memory used (and then streamed to a lagging follower) might be way bigger than the size of the snapshot itself (even by order of magnitude).

[marco6]

This PR lacks testing for now.

[codecov]

Codecov Report

Attention: Patch coverage is 80.51948% with 15 lines in your changes missing coverage. Please review.

Project coverage is 74.02%. Comparing base (36eac86) to head (3b87142).
Report is 11 commits behind head on master.

Files with missing lines	Patch %	Lines
src/raft/replication.c	53.33%	2 Missing and 5 partials ⚠️
src/server.c	70.00%	3 Missing ⚠️
src/raft/raft.c	75.00%	2 Missing ⚠️
test/raft/integration/test_snapshot.c	94.11%	0 Missing and 2 partials ⚠️
src/raft/log.c	90.00%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #714   +/-   ##
=======================================
  Coverage   74.02%   74.02%           
=======================================
  Files         192      192           
  Lines       27827    27899   +72     
  Branches     2797     2802    +5     
=======================================
+ Hits        20598    20653   +55     
- Misses       4906     4914    +8     
- Partials     2323     2332    +9     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[letFunny]

Thanks Marco for this PR, it looks very good to me, no comments so far. For the testing part of the story, we have the snapshot tests at test/raft/integration/test_snapshot.c so we could probably add this alternate strategy as a parameter for the existing tests there (where it makes sense). There are other tests that might be applicable such as the replication tests (test_replication.c).

We should also add a new test for the dynamic strategy asserting that the entries in the snapshot match the calculation for the expected size. Actually I was going to suggest changing existing tests to use both strategies by abstracting it away but I see that most of them use a threshold of 1 so it will probably require changing all of them, so adding a new one seems easier. I would suggest adding the tests in the unit tests for replication and in the integration tests, either in the replication_test or in snapshot_test.

Lastly, do not look at the tests under test/raft/unit/test_snapshot.c because those are for the new snapshot procedure that was never released.

[marco6]

After the conversation I had with @just-now, I changed slightly the strategy so that the computed trailing always sits in the range [threshold, trailing] so that the previous snapshot is still useful.

I also added some tests which hopefully cover the functionality now. I'm marking this as ready.

[just-now]

Looks very good. Just minor nit-picks.

[just-now]

Looks good. Merge it (preferably on Monday).