Poll node in batch to get validator indices

[twoeths]

Motivation

• Imported 160 oononba validators into contabo-1 and it failed to request validator indices due to the query max length

Description

• Query in batches, max batch=5
• Each batch contains at most 5 http requests to the node
• Each http request could work on 40 validators at most
• Work on at most 1000 validator pubkeys per poll

Let me know if we want to tweak the above numbers.

Closes #2727

Tested in contabo-1

Jun-22 09:24:26.394 []                 info: Decrypted 160 validator keystores
Jun-22 09:24:26.459 []                 info: Genesis available
Jun-22 09:24:26.477 []                 info: Verified node and validator have same config
Jun-22 09:24:26.727 []                 info: Discovered new validators count=160
Jun-22 09:24:36.614 []                 info: Published SyncCommitteeSignature slot=145718, count=160
Jun-22 09:24:36.832 []                 info: Published attestations slot=145718, index=0, count=5
Jun-22 09:24:40.068 []                 info: Published SyncCommitteeContribution slot=145718, index=1, count=1

[codeclimate]

Code Climate has analyzed commit 906d49c and detected 0 issues on this pull request.

View more on Code Climate.

[dapplion]

Looks good! What numbers do Teku use? We could base ours on them

[twoeths]

Looks good! What numbers do Teku use? We could base ours on them

@dapplion it's 10 per batch https://github.com/ConsenSys/teku/blob/79e49c48e0a6a78eac5b7760860d680a046e78af/validator/remote/src/main/java/tech/pegasys/teku/validator/remote/RemoteValidatorApiHandler.java#L75

I think it's quite conservative there, I used to be able to query 80 validator indices successfully, the issue only happened with 160 validators per batch. So

• 10 as Teku
• or 40 as in this PR
• or 20 which is in the middle

[wemeetagain]

Teku may be going off of this: https://stackoverflow.com/questions/3091485/what-is-the-limit-on-querystring-get-url-parameters#answer-56527673 I've seen this answer copy/pasted around, tho I can't find the original

Also related, I just posted this, maybe this can be more definitely resolved at the standards level: ethereum/beacon-APIs#153

[dapplion]

typo maxRequestsPerBatch

[dapplion]

Can find a similar approach without mutating scoped variables? You could have two for loops and scope the arrays inside them only with const

[github-actions]

Performance Report

✔️ no performance regression detected

Full benchmark results

Benchmark suite	Current: 6c3780c	Previous: 15766f4	Ratio
Process block - 250000 vs - with 0 validator exit	6.2893 ms/op	5.5543 ms/op	1.13
Process block - 250000 vs - with 1 validator exit	24.931 ms/op	22.065 ms/op	1.13
Process block - 250000 vs - with 16 validator exits	25.894 ms/op	22.641 ms/op	1.14
epoch - 250000 vs - processJustificationAndFinalization	101.72 us/op	89.589 us/op	1.14
epoch - 250000 vs - processRewardsAndPenalties	373.38 ms/op	329.08 ms/op	1.13
epoch - 250000 vs - processRegistryUpdates	9.9070 us/op	8.7800 us/op	1.13
epoch - 250000 vs - processSlashings	3.8138 ms/op	3.2909 ms/op	1.16
epoch - 250000 vs - processFinalUpdates	43.563 ms/op	37.574 ms/op	1.16
epoch - 250000 vs - prepareEpochProcessState	1.0065 s/op	876.69 ms/op	1.15
getAttestationDeltas - 250000 vs	168.65 ms/op	157.14 ms/op	1.07
processSlots - 250000 vs - 32 empty slots	1.9861 s/op	1.7430 s/op	1.14
shuffle list - 512 els	1.5292 ms/op	1.3357 ms/op	1.14
shuffle list - 16384 els	17.974 ms/op	15.492 ms/op	1.16
shuffle list - 250000 els	260.80 ms/op	226.70 ms/op	1.15
aggregationBits - 2048 els - readonlyValues	267.36 us/op	220.88 us/op	1.21
aggregationBits - 2048 els - zipIndexesInBitList	52.675 us/op	46.689 us/op	1.13
ssz.Root.equals	1.4670 us/op	1.1940 us/op	1.23
ssz.Root.equals with valueOf()	1.8490 us/op	1.6820 us/op	1.10
byteArrayEquals with valueOf()	1.7400 us/op	1.6680 us/op	1.04

by benchmarkbot/action