remove some deepcopy to speed up workflow conductor

[guzzijones]

1. removing deepcopy where we can
2. change deepcopy where still needed to custom copy command to increase speed using instead a shallow copy
3. add some data classes * tsc meeting comment

[guzzijones]

The use of 'latest' is causing my tox unit testing to fail it appears.

  orquesta.exceptions.PluginFactoryError: Unable to load plugin orquesta.composers.native. No 'orquesta.composers' driver found, looking for 'native'

I also had to disable python 3.6 for now.

[guzzijones]

@m4dcoder can you please take a look at this python 3.8 test failing. It looks like stevedore is not able to find the native, mock plugins to run the tests in python 3.8.

[guzzijones]

I did some testing with st2 and this definitely breaks with-items no longer report failure to the parent workflow.

(virtualenv) [vagrant@stackstorm]~/st2/tools$ st2 execution get 64adc15887cff69baae67294
id: 64adc15887cff69baae67294
action.ref: test_pack.test_fail_workflow
parameters:
  number: 5
status: running (635s elapsed)
start_timestamp: Tue, 11 Jul 2023 20:53:44 UTC
end_timestamp:
log:
  - status: requested
    timestamp: '2023-07-11T20:53:44.168000Z'
  - status: scheduled
    timestamp: '2023-07-11T20:53:44.245000Z'
  - status: running
    timestamp: '2023-07-11T20:53:44.317000Z'
result:
  output: null
+--------------------------+------------------------+-----------+---------------------+-------------------------------+
| id                       | status                 | task      | action              | start_timestamp               |
+--------------------------+------------------------+-----------+---------------------+-------------------------------+
| 64adc1581982b0a045f91e70 | succeeded (0s elapsed) | do_action | test_pack.test_fail | Tue, 11 Jul 2023 20:53:44 UTC |
| 64adc1591982b0a045f91e85 | succeeded (0s elapsed) | with_test | test_pack.test_fail | Tue, 11 Jul 2023 20:53:45 UTC |
| 64adc1591982b0a045f91e88 | succeeded (1s elapsed) | with_test | test_pack.test_fail | Tue, 11 Jul 2023 20:53:45 UTC |
| 64adc1591982b0a045f91e8b | failed (1s elapsed)    | with_test | test_pack.test_fail | Tue, 11 Jul 2023 20:53:45 UTC |
| 64adc1591982b0a045f91e8f | succeeded (1s elapsed) | with_test | test_pack.test_fail | Tue, 11 Jul 2023 20:53:45 UTC |
| 64adc1591982b0a045f91e92 | succeeded (1s elapsed) | with_test | test_pack.test_fail | Tue, 11 Jul 2023 20:53:45 UTC |
+--------------------------+------------------------+-----------+---------------------+-------------------------------+

[guzzijones]

We won't be able to add a feature flag without a change in ST2. Orquesta does not have the Stackstorm config object afaik.
I will put some benchmarks in the PR and also make sure we double check that errors is also copied.

[Kami]

I will put some benchmarks in the PR and also make sure we double check that errors is also copied.

Thanks.

We won't be able to add a feature flag without a change in ST2. Orquesta does not have the Stackstorm config object afaik.

Yeah, I assume it would probably also require a change on the StackStorm side. If you have cycles, I still think that would be a good idea.

Still much easier to flip the feature flag / config option in case something goes wrong versus needing to downgrade to a previous release.

Especially if it's combined with a breaking backward incompatible change in StackStorm core (live action FK change - there is only a forward migration in that PR, but not a backward one).

[rebrowning]

I've been looking into some slowness for workflows with a large amount of items inside of a with items step, I saw this PR and I'm curious what items are we actually looking to have gated by a feature flag in this PR?

[guzzijones]

task render speed gain when removing deep copy for large parameters

[guzzijones]

I added a micro benchmark to show how much faster it is to NOT do a deep copy. I could add it for the serialize and deserialize, but the results will be the same. Copying large objects is costly.

[guzzijones]

all the 3.9 tests passed with these updates minus the lint checks).
rpm is here if @rebrowning wants to test it out.

[guzzijones]

I am running this additional speed up on our internal fork of St2.
I will report back if I see any issues.

It looks like ci/cd failed because black changed with their latest release.

[rebrowning]

@Kami @guzzijones are we requiring the feature flag to switch between deepcopy and fastdeepcopy for this to move forward?

[FileMagic]

Note, this will help with "with items" speed issues.

@Kami is not active on this project anymore it seems

[guzzijones]

@nzlosh @cognifloyd please take a peak at this. I adds significant speed improvements by eliminating some deep copies. This helps with larger context objects.

[guzzijones]

Anyone willing to give some feedback here?

[nzlosh]

Technically I'm OK with this to be merged. As discussed mid 2024, it should be merged as part of st2 3.10. I think we should leave a release between the pymongo updates and the deepcopy change to avoid confusing any issues that may come up in the 3.9 and 3.10 release.