Stream job results in chunks

[Mandrenkov]

Context:

Currently, fetching large job results (e.g., those on the order of GBs) can take an unreasonably long time due to the memory overhead associated with the default fetching strategy of the requests library. Fetching the same job result from a web browser (e.g., Firefox) tends to be much faster presumably due to the streaming nature of the download.

Description of the Change:

• Modified Connection.request() to respect the stream keyword argument during error handling.
• Modified Job.result to stream job results from the Xanadu Cloud platform in chunks.

Benefits:

• Large job results are fetched significantly faster, even more so as the size of the job result grows.

Possible Drawbacks:
None.

Related GitHub Issues:
None.

[doctorperceptron]

Looks good! Since everything is now streamed I'm assuming that streaming small files isn't any slower than before?

[Mandrenkov]

Thanks for the review, @doctorperceptron!

Looks good! Since everything is now streamed I'm assuming that streaming small files isn't any slower than before?

Excellent question! To check if this is the case, I submitted a small job and fetched the results 10 times using this branch as well as main. For reference, I submitted the job using

xcc job submit simulon_gaussian "name x\nversion 1.0\ntarget simulon_gaussian (shots=2)\nMeasureFock() | [0, 1, 2, 3]"

The results show there is no significant difference in the fetching time of this job between the branches.

Branch main

            Mean        Std.Dev.    Min         Median      Max
real        1.507       0.175       1.152       1.532       1.716
user        0.573       0.051       0.508       0.569       0.710
sys         0.694       0.045       0.582       0.697       0.767

Branch sc-17551-stream-job-results

            Mean        Std.Dev.    Min         Median      Max
real        1.595       0.219       1.075       1.611       1.993
user        0.508       0.033       0.445       0.515       0.545
sys         0.682       0.032       0.627       0.679       0.735

[doctorperceptron]

Sounds good! 🚀