Pulses debug mode

[alecandido]

Pulses are now stored with IDs, but there are cases in which the ID would mask too much of the information, and we need a way to easily retrieve the identity of the pulse.

My idea is to accompany pulses with a dump of their registry.

So, every time that we execute a sequence, we should propagate the information about the input. This could be done:

• serializing the sequence on a log file
• propagating a handle to the sequence inside the drivers, such that the information is always aside the ID

I'm a bit against the second option, but it could make someone's life easier.

Instead, for the first one we could have a CLI, or some script, to handle a sequence storage, such that given a sequence ID and pulse ID, you could always retrieve enough information to reconstruct the whole pulse (or the whole sequence).

$ qibolab pulse 12903470.1203947
{
  "frequency": ...,
  ...
}
$ qibolab pulse 12398412
[{ ... }, { ... }, ...]

and in python this could be the JSON of the dict obtained from asdict(pulse), [asdict(p) for p in sequence].

We could just avoid logging by default (to have a disk-free production mode), and if configured with a suitable path instead log every sequence that will be executed (or played, according if we want to do it at platform or controller level).

[alecandido]

@stavros11

[stavros11]

Maybe I am missing something, but is the proposal to dump a mapping from id(pulse) to pulse parameters asdict(pulse) to disk? For debugging QUA scripts this would work, it would just be a bit more annoying as we would have to dump two files (qua script + pulse log) and match ids between the two.

[alecandido]

but is the proposal to dump a mapping from id(pulse) to pulse parameters asdict(pulse) to disk?

Yes, and additional utilities on that base.

E.g. for QUA we could make a simple (external) tool that does the replacement for you.

[stavros11]

E.g. for QUA we could make a simple (external) tool that does the replacement for you.

Indeed, that works for me. I just have two suggestions for the actual implementation:

• include an identifier in the pulse log file, both inside the json and even in the file name, that shows which execution that produced that log. This could be a simple timestamp. I think it is relevant because the ids are different in every execution anyway.
• control the debug mode through environment variable, so that it is possible to turn on/off without touching the code we are debugging. For pulses we could have something like

debug_path = os.environ.get("QIBOLAB_DEBUG", None)
if debug_path is not None:
    sequence.dump(debug_path)

inside platform.execute, to also allow the user to select where to dump the logs (and we can use the same path in the drivers if applicable).

[alecandido]

• include an identifier in the pulse log file, both inside the json and even in the file name, that shows which execution that produced that log. This could be a simple timestamp. I think it is relevant because the ids are different in every execution anyway.

I was considering doing this anyhow, but thinking a bit more I'm not sure it is actually needed: using UUIDs for pulses they will always be unique, also across runs (they essentially contain a timestamp). So, in principle, we could even store all of them in a single file (it will just take more to load, if used a lot).

• control the debug mode through environment variable, so that it is possible to turn on/off without touching the code we are debugging. For pulses we could have something like

Agreed: as said above, you do not want to dump in production, and the env var might be convenient to avoid dedicated scripts for testing.

[alecandido]

Internal feature, postponed to after 0.2.0 release.