use monotonic-timestamp in create()

[staltz]

ssb-db and ssb-validate both use monotonic-timestamp to avoid the problem where two consecutive calls of Date.now() would give the same number. In our case this would help so we can avoid the Date.now() + 1, Date.now() + 2 hacks. I do think that the async-await we use in create() already guarantees that Date.now() will be different (I was unable to create a 1st :x: 2nd :heavy_check_mark: test), but it makes sense to anyway use monotonic-timestamp to be sure we never bump into this problem.

[staltz]

I'm not sure what's going on with the tests, they pass locally, but we definitely need to fix them in CI before merging.

[staltz]

The reason why tests were broken was that monotonic-timestamp was being replaced by mock-monotonic-timestamp (because the way we use ssb-fixtures ends up "leaking" this hack globally) so that there was a test that checked that the msg.timestamp should be greater than some Date.now(), but the msg.timestamp was in the year 2015, because of mock-monotonic-timestamp.

I used pnpm in CI, which fixed that error, but then other problems came up: some timestamps are floats, like 1659105426666.001 which doesn't work in bendybutt because ssb-bfe only bypasses integers, but also it turns out that bencode only supports integers, not floats. The JS library bencode will convert the float to an integer.

@arj03 What do you think, should we quit this PR?

UPDATE: I guess the tests passed this time randomly, but we're sure that sometimes we'll get a float timestamp and bendybutt will break.

[github-actions[bot]]

Benchmark results

Part	Duration
Create 5000 new messages	560.51ms
Validate 5000 messages	639.76ms
Native to db format 5000 messages	153.66ms
Db to native format 5000 messages	118.68ms
Add 1000 elements	542.95ms
Add 1000 box1 msgs	1326.61ms
Unbox 1000 box1 msgs first run	246.76ms
Unbox 1000 box1 msgs second run	143.90ms
Add 1000 box1 msgs	1274.90ms
Query 1000 msgs first run	47.61ms
Query 1000 msgs second run	29.09ms
Add 1000 box2 msgs	1904.28ms
Unbox 1000 box2 msgs first run	326.09ms
Unbox 1000 box2 msgs second run	247.98ms
Migrate (+db1)	16266.35ms
Migrate (alone)	6052.15ms
Migrate (+db1 +db2)	12551.21ms
Migrate (+db2)	8632.67ms
Migrate continuation (+db2)	1353.85ms
Memory usage without indexes	754.91 MB = 39.51 MB + etc
Initial indexing	885.96ms
Initial indexing maxcpu=86	4893.51ms
Initial indexing compat	1120.10ms
Two indexes updating concurrently	1351.57ms
Key one initial	52.87ms
Key two	1.52ms
Key one again	1.38ms
Reboot and key one again	66.90ms
Latest root posts	1086.30ms
Latest posts	15.28ms
Votes one initial	725.03ms
Votes again	1.15ms
HasRoot	494.84ms
HasRoot again	0.37ms
Author one posts	644.45ms
Author two posts	21.04ms
Dedicated author one posts	641.23ms
Dedicated author one posts again	0.70ms
DeleteFeed	3957.05ms
Maximum memory usage	1029.91 MB = 65.60 MB + etc
Indexes folder size	10.01mb

[arj03]

Regarding bendy butt I guess it's a matter of interpretation. I really dislike that we allowed floating timestamps in classic and think it would be better overall to convert all of them to int.

[staltz]

So does that mean we should make a new (version or fork) monotonic-timestamp that only gives integers?

[arj03]

That would make this PR a lot more tasty :)

[staltz]

Hmm, pretty sure we could rewrite a new PR instead of working on this old one. I want to kill some old PRs.