# fusaka-bugs
Here is a collection of overall bugs across devnet-3, devnet-5 and mainnet:
**Prysm**:
During the latest non finality, we had an issue when starting a node from a checkpoint with a pre-existing DB.
(Starting from genesis was OK, and starting from a checkpoint with an empty DB was OK.
This bug was possibly more likely to happen when the latest justified checkpoint was far away from the latest finalized checkpoint.
(When, for example, we have a single checkpoint with > 2/3 participation).
**Teku**:
OOM bug: https://github.com/Consensys/teku/issues/9824
Syncing bug: https://github.com/Consensys/teku/issues/9826
Upgrade incident bug (affecting mainnet): https://hackmd.io/@teku/B1lVi-Scxg
OOM bug mainly triggered by malicious node not getting rate limited.
Example:
16Uiu2HAmHbojWU7RFGhFGpU1rVhcJJRjfffR5pZSF7RAojhwhNvv
it was constantly asking 512 columns by root. Leading to huge mem spike and crashing teku nodes
CKZG bug with the alpha.6 specs on 2.1.2 as well as 2.1.3 (needed cellids sorted as input, fixed in client)
Prometheus lib returning broken pipe
**Lighthouse**:
issues maintaining peers across all custody subnets. Fix is in the works
issues syncing on “forky” chains. Fix in the works. The fix should make syncing more stable in general
validator overwhelming the BN with requests while the node isn’t synced (resolved)
**Lodestar**: This is the relevant stuff that got merged after the non-finality event. Some of this though is not specifically pertinent but all is tangential to sync and peer scrore to maintain a mesh of peers to pull from
- [increase max peers to 200 for better peer pool to request from](https://github.com/ChainSafe/lodestar/pull/8272)
- [fix how we deserialize data from getBlobsV2 to drastically improve performance](https://github.com/ChainSafe/lodestar/pull/8275)
- [self rate limiter to handle dead requests](https://github.com/ChainSafe/lodestar/pull/8295)
- [serve data columns by root for finalized epochs](https://github.com/ChainSafe/lodestar/pull/8250)
- our data column database is broken right now, we are likely fetching the wrong column as our indexing is wrong, will have a fix soon
**Nimbus**:
- a column reconstruction issue where it wasn't filling in all fields, so was making invalid columns
- sync initially didn't know how to find/filter for peers with relevant columns and could get stuck
- using 8 columns without validators
**Grandine**:
we suffered from sync speed, and data redownload loop from the last finalized slot when it retry download many times but can't move local head. this is the expected behavior with our syncing strategy, the latest changes could partially fix this issue. but even though the network has been recovered (finalized again), the finalized slot still can't move as we can see in grandine-geth-1 instance, so I think it is a potential bug
**Reth**:
Incorrect blob fee comparison: https://github.com/paradigmxyz/reth/pull/18216
mainnet bug: https://github.com/paradigmxyz/reth/issues/18205
Reth failed to build blocks on commit `3ce0a381085bc0547136c41577801aea1e7f47d9`, fixed in latest main
**Erigon**:
We have still some issues with erigon. See Eth R&DErigon <> Prysm.
Errors like:
- [2025-09-01 08:14:14.42] ERROR execution: Got a validation error in newPayload error=invalid block[5/5 Execution] wrong trie root
- [2025-08-28 17:17:14.38] ERROR sync: Could not reconstruct full bellatrix block batch from blinded bodies error=received 0 payload bodies from engine_getPayloadBodiesByRangeV1 with count=3 (start=211195): engine api payload body response is invalid
`block has invalid execution payload` related to reorg bug
Usually wiping everything fixes the issue.
**NimbusEL**: Reported OOM issues being investigated.
**Besu** : [Report](https://hackmd.io/@siladu/ryrMTURcex)
[devnet-5 issues](https://hackmd.io/@siladu/rkePw-Rill)
**Geth**:
* CRIT [09-04|14:15:15.182] Failed to truncate extra state histories err="history head truncation out of range, tail: 154748, head: 244349, target: 244748"
Bug relating to corrupt db when started at genesis after restarts
Snap syncing bug, full syncing continued to work fine
**Nethermind**:
Deep reorg issues: A reorg has led to picking a parent block from a different branch, which was declined by internal checks and the system got stuck. It has been reproduced locally and the fix is on the way, just need to test it more