op-batcher: Move decision about data availability type to channel submission time

[geoknee]

Closes #11609

For now, I decided to not make any changes to the decision logic itself (only to when it is triggered and what happens when a change is necessary).

Changing the trigger for the decision opens the opportunity to make a more accurate decision about which DA type to choose, since some transaction data is now in scope. However, this would not address the larger assumption in the current logic, which is about how many blobs are typically included in a tx. This can be lower than the target number on chains with low throughput as channels must be closed before they time out. To address this assumption, we could store the previous number of blobs per tx to make a better estimate than the current one (which assumes we always include the target number). This is saved for future work.

Tests: the modified implementation passes existing tests, including an end to end test for switching DA type.

[sebastianst]

We should definitely add a few unit tests of TxData() and the requeue function (currently called Rebuild).

[zhiqiangxu]

If the Data Availability type is determined at submission time, how are MaxFrameSize and MaxFramesPerTx decided? I believe there's a kind of interdependency here.

[geoknee]

If the Data Availability type is determined at submission time, how are MaxFrameSize and MaxFramesPerTx decided? I believe there's a kind of interdependency here.

In dynamic DA mode, the batcher maintains two configs (one for each DA type blobs and calldata). One of those configs is the default (i.e. initially blobs, but generally whatever was used when the previous channel was submitted). The channel can then be built optimistically with the default config. When it comes to submit the channel, if the data availability type needs to change, channels which haven't been submitted will be discarded and rebuilt with the appropriate config (including params like MaxFrameSize. Does that make sense to you?

[zhiqiangxu]

If the Data Availability type is determined at submission time, how are MaxFrameSize and MaxFramesPerTx decided? I believe there's a kind of interdependency here.

In dynamic DA mode, the batcher maintains two configs (one for each DA type blobs and calldata). One of those configs is the default (i.e. initially blobs, but generally whatever was used when the previous channel was submitted). The channel can then be built optimistically with the default config. When it comes to submit the channel, if the data availability type needs to change, channels which haven't been submitted will be discarded and rebuilt with the appropriate config (including params like MaxFrameSize. Does that make sense to you?

Yeah it makes sense, thank you!

[sebastianst]

In dynamic DA mode, the batcher maintains two configs (one for each DA type blobs and calldata). One of those configs is the default (i.e. initially blobs, but generally whatever was used when the previous channel was submitted). The channel can then be built optimistically with the default config. When it comes to submit the channel, if the data availability type needs to change, channels which haven't been submitted will be discarded and rebuilt with the appropriate config (including params like MaxFrameSize. Does that make sense to you?

Worth adding that we're effectively still choosing the config at the same point in time when we create the channel. That's because right after a channel is being submitted, a new channel is created, and then we choose the config that was just selected 2 seconds ago.

[sebastianst]

thanks for adding the additional tests!

[semgrep-app]

Semgrep found 1 sol-style-return-arg-fmt finding:

• packages/contracts-bedrock/src/libraries/Blueprint.sol
  • L59 - Triage

Named return arguments to functions must be appended with an underscore (_)

_{Ignore this finding from sol-style-return-arg-fmt.}

[semgrep-app]

Semgrep found 2 sol-style-input-arg-fmt findings:

• packages/contracts-bedrock/scripts/DeployImplementations.s.sol
  • L473-476 - Triage
  • L769-772 - Triage

Inputs to functions must be prepended with an underscore (_)

_{Ignore this finding from sol-style-input-arg-fmt.}