Prater sync failure with invalid public key.

[mcdee]

Description

Teku stops syncing on goerli/prater at slot 3047065. Truncated error logs are:

Caused by: tech.pegasys.teku.bls.impl.BlsException: Deserialization of public key bytes failed: 0x000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
at tech.pegasys.teku.bls.impl.blst.BlstPublicKey.fromBytes(BlstPublicKey.java:43) ~[teku-infrastructure-bls-22.8.2.jar:22.8.2]
...
Caused by: java.lang.RuntimeException: BLST_ERROR: bad point encoding
at supranational.blst.blstJNI.new_P1_Affine__SWIG_1(Native Method) ~[jblst-0.3.8.jar:?]
...
2022-09-01 00:00:23.263 WARN  - Failed to process block
tech.pegasys.teku.bls.impl.BlsException: Failed to fastAggregateVerify
at tech.pegasys.teku.bls.BLS.fastAggregateVerify(BLS.java:203) ~[teku-infrastructure-bls-22.8.2.jar:22.8.2]
...
2022-09-01 00:00:23.264 WARN  - State transition error while importing block 3047065 (0xb1e4fad4c8b6042ecab18f6e0fbd815d8f3d69496837132dc2b69763323cd4e6)

Full log after a restart at teku.log.gz

Steps to Reproduce (Bug)

This is reproducible on the given instance on restart, however I have not seen this on any other instance (either prater or other networks) so is not reproducible outside of this one instance (db available on request).

Versions (Add all that apply)

Teku 22.8.2

[asanso]

@ajsutton @mcdee was this fixed or why was it closed ?

[benjaminion]

I don't have full context any more, but the timing seems consistent with this PR - likely there was a bug that we quietly fixed via that reversion without advertising its existence. The others can comment if they can still remember.

[ajsutton]

Yeah Ben's spot on. The BLS library changed from returning false when verification failed to throwing an exception which caused the issue.