- Add new command to the cli/node: `serveapi` that alows serving the API just
by connecting to the PostgreSQL database. The mode flag should me passed in
order to select whether we are connecting to a synchronizer database or a
coordinator database. If `coord` is chosen as mode, the coordinator
endpoints can be activated in order to allow inserting l2txs and
authorizations into the L2DB.
Summary of the implementation details
- New SQL table with 3 columns (plus `item_id` pk). The table only contains a
single row with `item_id` = 1. Columns:
- state: historydb.StateAPI in JSON. This is the struct that is served via
the `/state` API endpoint. The node will periodically update this struct
and store it int he DB. The api server will query it from the DB to
serve it.
- config: historydb.NodeConfig in JSON. This struct contains node
configuration parameters that the API needs to be aware of. It's updated
once every time the node starts.
- constants: historydb.Constants in JSON. This struct contains all the
hermez network constants gathered via the ethereum client by the node.
It's written once every time the node starts.
- The HistoryDB contains methods to get and update each one of these columns
individually.
- The HistoryDB contains all methods that query the DB and prepare objects that
will appear in the StateAPI endpoint.
- The configuration used in for the `serveapi` cli/node command is defined in
`config.APIServer`, and is a subset of `node.Config` in order to allow
reusing the same configuration file of the node if desired.
- A new object is introduced in the api: `StateAPIUpdater`, which contains all
the necessary information to update the StateAPI in the DB periodically by
the node.
- Moved the types `SCConsts`, `SCVariables` and `SCVariablesPtr` from
`syncrhonizer` to `common` for convenience.
In the metrics query, the estimatedTimeToForgeL1 was probably not being read
from the SQL query because it contained upper case letters, and in PostgreSQL
when identifiers are not double quoted they are case insensitive:
https://stackoverflow.com/a/20880247
Previously the code was only querying the unforged L1UserTxs of a particular
queue, but this was incorrect because there are always two non-forged queues:
the frozen one and the open one.
Replace it by a query to all the unforged L1UserTxs via a new HistoryDB method.
- Add config parameter `Coordinator.L2DB.MinPriceUSD` which allows rejecting
txs to the pool that have a fee lower than the minimum.
- In pool tx insertion, checking the number of pending txs atomically with the
insertion to avoid data races leading to more than MaxTxs pending txs in the
pool.
- cli / node
- Update handler of SIGINT so that after 3 SIGINTs, the process terminates
unconditionally
- coordinator
- Store stats without pointer
- In all functions that send a variable via channel, check for context done
to avoid deadlock (due to no process reading from the channel, which has
no queue) when the node is stopped.
- Abstract `canForge` so that it can be used outside of the `Coordinator`
- In `canForge` check the blockNumber in current and next slot.
- Update tests due to smart contract changes in slot handling, and minimum
bid defaults
- TxManager
- Add consts, vars and stats to allow evaluating `canForge`
- Add `canForge` method (not used yet)
- Store batch and nonces status (last success and last pending)
- Track nonces internally instead of relying on the ethereum node (this
is required to work with ganache when there are pending txs)
- Handle the (common) case of the receipt not being found after the tx
is sent.
- Don't start the main loop until we get an initial messae fo the stats
and vars (so that in the loop the stats and vars are set to
synchronizer values)
- When a tx fails, check and discard all the failed transactions before
sending the message to stop the pipeline. This will avoid sending
consecutive messages of stop the pipeline when multiple txs are
detected to be failed consecutively. Also, future txs of the same
pipeline after a discarded txs are discarded, and their nonces reused.
- Robust handling of nonces:
- If geth returns nonce is too low, increase it
- If geth returns nonce too hight, decrease it
- If geth returns underpriced, increase gas price
- If geth returns replace underpriced, increase gas price
- Add support for resending transactions after a timeout
- Store `BatchInfos` in a queue
- Pipeline
- When an error is found, stop forging batches and send a message to the
coordinator to stop the pipeline with information of the failed batch
number so that in a restart, non-failed batches are not repated.
- When doing a reset of the stateDB, if possible reset from the local
checkpoint instead of resetting from the synchronizer. This allows
resetting from a batch that is valid but not yet sent / synced.
- Every time a pipeline is started, assign it a number from a counter. This
allows the TxManager to ignore batches from stopped pipelines, via a
message sent by the coordinator.
- Avoid forging when we haven't reached the rollup genesis block number.
- Add config parameter `StartSlotBlocksDelay`: StartSlotBlocksDelay is the
number of blocks of delay to wait before starting the pipeline when we
reach a slot in which we can forge.
- When detecting a reorg, only reset the pipeline if the batch from which
the pipeline started changed and wasn't sent by us.
- Add config parameter `ScheduleBatchBlocksAheadCheck`:
ScheduleBatchBlocksAheadCheck is the number of blocks ahead in which the
forger address is checked to be allowed to forge (apart from checking the
next block), used to decide when to stop scheduling new batches (by
stopping the pipeline). For example, if we are at block 10 and
ScheduleBatchBlocksAheadCheck is 5, eventhough at block 11 we canForge,
the pipeline will be stopped if we can't forge at block 15. This value
should be the expected number of blocks it takes between scheduling a
batch and having it mined.
- Add config parameter `SendBatchBlocksMarginCheck`:
SendBatchBlocksMarginCheck is the number of margin blocks ahead in which
the coordinator is also checked to be allowed to forge, apart from the
next block; used to decide when to stop sending batches to the smart
contract. For example, if we are at block 10 and
SendBatchBlocksMarginCheck is 5, eventhough at block 11 we canForge, the
batch will be discarded if we can't forge at block 15.
- Add config parameter `TxResendTimeout`: TxResendTimeout is the timeout
after which a non-mined ethereum transaction will be resent (reusing the
nonce) with a newly calculated gas price
- Add config parameter `MaxGasPrice`: MaxGasPrice is the maximum gas price
allowed for ethereum transactions
- Add config parameter `NoReuseNonce`: NoReuseNonce disables reusing nonces
of pending transactions for new replacement transactions. This is useful
for testing with Ganache.
- Extend BatchInfo with more useful information for debugging
- eth / ethereum client
- Add necessary methods to create the auth object for transactions manually
so that we can set the nonce, gas price, gas limit, etc manually
- Update `RollupForgeBatch` to take an auth object as input (so that the
coordinator can set parameters manually)
- synchronizer
- In stats, add `NextSlot`
- In stats, store full last batch instead of just last batch number
- Instead of calculating a nextSlot from scratch every time, update the
current struct (only updating the forger info if we are Synced)
- Afer every processed batch, check that the calculated StateDB MTRoot
matches the StateRoot found in the forgeBatch event.
L2Tx.TokenID is not on the data obtained by the Synchronizer from the
blockchain, but is set by the TxProcessor when processing the
transactions in the StateDB.
Previously the coordinator was erroneously using Slot.BatchesLen to determine
when anyone can forge. The correct behaviour is implmenented with the boolean
flag `ForgerCommitment`, that is set to true only when there's a batch before
the deadline within a slot.
Delete Slot.BatchesLen, and the synchronization code of this value from the
Synchronizer, as this is not needed
Update usage of `*babyjub.PublicKey` to `babyjub.PublicKeyComp`
- when the key is not defined, internally is used `babyjub.EmptyBJJComp`, which is a `[32]byte` of zeroes of type `babyjub.PublicKeyComp`
- the API continues returning `nil` when the key is not defined
When the synchronizer queries the unforged L1UserTxs, sort them by position
Swap the order of calls setL1UserTxEffectiveAmounts and addBatch in
AddBlockSCData because otherwise, for reasons I dont understand, the item_id of
the txs doesn't follow the position of the txs.
For each tx, move the logic of setting the Type and TxID to separate functions,
so that they can be called when necessary.
In synchronizer, set all the required fields using the `SetID` and `SetType`
for l2txs when needed. This is necessary because the `ProcessTxs()` works with
`common.PoolL2Tx`, which misses some fields from `common.L2Tx`, and because
`ProcessTxs()` needs the Type to be set, but at the same time `ProcessTxs()`
sets the Nonce, which is required for the `TxID`.
Eduard S
arnaubennassar
Danilo Pantani
arnaucube
Toni Ramírez
laisolizq