ethdb/pebble: upgrade pebble

[rjl493456442]

This PR upgrades the pebble dependencies with two features merged:

• Check bloom filter in Get function cockroachdb/pebble#2846
• Implement APIs for creating batch with pre-allocated size cockroachdb/pebble#2828

The intention is not to merge this pull request immediately, but wait a bit more for a stable release from pebble.

[rjl493456442]

Run a snap sync on bench05/06, 05 is the experimental branch.

There is no big difference of the overall system resource comsumption.

Database read/write wise, master is "slightly" higher. But we can treat them as identical.

05 finishes sync a half hour ahead (9 hours in total), not too much difference.

All in all, I don't think pebble changes has a big impact on the snap sync performance. Let's switch to full sync then.

[rjl493456442]

https://geth-bench.ethdevops.io/d/Jpk-Be5Wk/dual-geth?var-exp=bench05&var-master=bench06&var-percentile=50&from=now-15m&to=now&refresh=1m

Start another round of benchmark, run full sync on bench05/06, 05 is PR.

[MariusVanDerWijden]

LGTM

[rjl493456442]

Switch to full sync for a few days, dump some metrics here

Overall system metrics

Not too much difference

Database

This pull request has ~10% disk read, otherwise pretty much same(bench05 is 4 hours ahead)

Sync performance

This pull request is ~3hours a head after running 3 or 4 days. A little faster, but not too much

The conclusion is, with the change of cockroachdb/pebble#2846, a few unnecessary disk reads are avoid.

[karalabe]

Given the strange IO issues on Pebble 23.1, this branch got a priority merge 🙃