checkpointing broken in lightmode on mainnet

[sidhujag]

System information

Geth version: 1.9.5
OS & Version: OSX
Commit hash : stable

Expected behaviour

Mainnet to sync from hardcoded checkpoint in light mode

Actual behaviour

Mainnet in lightmode does not sync from hardcoded checkpoint, starts from 0 every time.

Steps to reproduce the behaviour

1. clear your geth blocks or data directory
2. ./geth --syncmode light

Backtrace (with verbosity enabled)

DEBUG[09-27|14:29:16.259|light/lightchain.go:379]            Inserted new header                      number=1       hash=88e96d…406cb6

Note that if you have a hardcoded checkpoint you should never see that message if checkpointing is enabled.

I get the:
DEBUG[09-27|14:29:01.460|les/sync.go:130] Disable checkpoint syncing reason="empty checkpoint"
message meaning it will disable checkpoint and fall back to lightSync mode:

go-ethereum/les/sync.go

Line 129 in 32b07e8

mode = lightSync

where it will completely disable checkpointing, hardcoded/oracle or whatever.

Note the fix is to simply post (hack imo) a checkpoint on the mainnet oracle https://etherscan.io/address/0x9a9070028361f7aabeb3f2f2dc07f82c4a98a02a#events which was created 80 days ago but no checkpoints posted. Rinkeby has checkpoints and thus even though the checkpoints are older than the hardcoded one, it still relies on the hardcoded checkpoint because of the fact that checkpoints response is not empty.

The proper fix likely is to re-validate the logic that you should probably fallback to legacy checkpoint mode if no oracle checkpoints exists (what you had before this bug existed).

[rjl493456442]

Could you please try the latest master? There was an initialization error before and is fixed on master now.

[sidhujag]

When is the next release? We usually roll up production versions with md5 hash announced on our end, you fixed in #20120 right? Looks ok, when is 1.9.6 going to release?

[rjl493456442]

@sidhujag Usually we have our biweekly release. In theory, we will have a new release in following 1 or 2 weeks. But it's Devcon very soon, not sure about it.

[rjl493456442]

@sidhujag Regarding the fix, yes it's 20120

[sidhujag]

@sidhujag Regarding the fix, yes it's 20120

ok please ping me when you have it and I will give an update on this ticket, enjoy Devcon!

[rjl493456442]

@sidhujag will do!

[rjl493456442]

@sidhujag 1.9.6 is released. Please check it. Close the issue now and feel free to reopen if issue still there.

[sidhujag]

Tested ACK