Summary
The user is facing an issue where newly registered launch plans created with LaunchPlan.get_or_create(...)
are not appearing connected to a workflow, despite being visible in the Flyte console. They suspect this problem started after a recent update to their Flyte deployment via Terraform. The new launch plan "z_detector" is visible in the Launch Plans tab but not in the Workflows tab, which is preventing the execution of updated versions of existing launch plans. The user is reviewing their deployment code for issues, and the problem persists even after deleting the launch plan, suggesting the workflow itself isn't updating. They note that this issue affects all workflows, with the last update for any workflow being on 8/26. The user later realizes that a recent refactor added a directory to the workflows' path, causing the problem. They plan to clear out old workflows to ensure only new ones are displayed and are trying to understand why old workflows are still running when triggering a workflow run.
ytong
ofc
andrew.merrill
for sure. thanks for the help! that helped me realize that
ytong
thanks! let us know
ytong
like a folder gets added and i kept searching for the old path.
ytong
i’ve run into that before
ytong
oh yeah, should’ve thought of that.
andrew.merrill
I just need to figure out why the old workflows are still running when I trigger a workflow run, but I think I can figure that out
andrew.merrill
yeah, it looks like that's the issue, i need to see if I can clear out these workflows so that only the new ones show up... that's my bad, it was just the directory path issue
andrew.merrill
oh man.. I think I see the issue. there was a refactor a little bit ago that added a directory in the path to the workflows, and now I'm seeing the new workflow with the new path showing up
ytong
and you’re sure this is the correct project/domain
ytong
and if you click on one… then manually change the version to Mxuji6SmjvFRo5-uGYS5yg
do you get a 404?
andrew.merrill
looks like its true for all of my workflows. the latest update for any workflow was 8/26. that's making me think it is some kind of deployment issue..
andrew.merrill
yeah, it isn't there either. the versions go back to january, but the august one is still the latest
ytong
and if you click on the View All… it’s still missing?
ytong
got it okay
andrew.merrill
yeah, I guess its more that the workflow itself isn't updating at all
ytong
the problem persists even if you delete this launch plan.
ytong
the issue is that the workflows page is missing the workflow that is registered.
ytong
so just to clarify… this doesn’t really have to do with the launch plan right?
andrew.merrill
I'm trying to look through my deployment code to see if anything ended up weird there, because its weird that it just stopped updating a month ago
andrew.merrill
Yeah, here's a modified version of the launchplan in the code
z_launch_plan = LaunchPlan.get_or_create(
name=z_launch_plan_name,
workflow=detector_pipeline,
default_inputs={
**default_inputs,
"epochs": 16,
"batch_size": 28,
},
fixed_inputs={
"model_type": "detector",
"model_category": "z",
},
)```
where `detector_pipeline` is the name of the workflow in the code
ytong
which workflow does z_detector_pipeline_launchplan
correspond to? detector.detector_pipeline.detector_pipeline
?
andrew.merrill
Yeah, here's the output
Registering against <http://flyte.domain.com|flyte.domain.com>
Detected Root path/to/code, using this to create deployable package...
No output path provided, using a temporary directory at /var/folders/fg/rj4k8n6n6gz2mvvsqdv34_1h0000gn/T/tmp_u9v4qpo instead
Computed version is Mxuji6SmjvFRo5-uGYS5yg
Loading packages ['detector.detector_pipeline'] under source root path/to/code
Image us-central1-docker.pkg.dev/dai-ml-pipelines/flyte-pipelines/default:QbhpPoJ1HnMIoJtST_44bg found. Skip building.
Image us-central1-docker.pkg.dev/dai-ml-pipelines/flyte-pipelines/gcloud:DHMde3nJznQF0DCOHcCFDA found. Skip building.
Image us-central1-docker.pkg.dev/dai-ml-pipelines/flyte-pipelines/detector_train:aeK6Nv9bo_qYB2HjeKq_FA found. Skip building.
Image us-central1-docker.pkg.dev/dai-ml-pipelines/flyte-pipelines/detector_serialize:URlgVNOdiYdTZ7reC_w7bA found. Skip building.
Successfully serialized 24 flyte objects
[✔] Registration common.tasks.pipeline_setup.get_pipeline_steps type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration common.tasks.pipeline_setup.get_dataset_bucket_dir type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration common.tasks.pipeline_setup.return_float type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration common.tasks.pipeline_setup.create_pipeline_status type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration common.tasks.data_validation.validate_dataset type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration detector.steps.detector_bigquery.setup_detector_bigquery type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration common.tasks.pipeline_setup.return_str type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration common.tasks.pipeline_setup.noop type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration detector.steps.detector_data_processing.process_detector_dataset type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration common.tasks.pipeline_setup.get_output_dir type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration common.tasks.pipeline_setup.return_str_float_tuple type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration common.tasks.deploy.trigger_model_deployment type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration detector.steps.detector_serialize.serialize_detector type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration detector.steps.detector_evaluate.evaluate_detector type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration detector.steps.detector_train.train_detector type TASK successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration detector.detector_pipeline.detector_train_eval_serialize type WORKFLOW successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration detector.detector_pipeline.detector_train_eval_serialize type LAUNCH_PLAN successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration detector.detector_pipeline.detector_pipeline type WORKFLOW successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration detector.detector_pipeline.detector_pipeline type LAUNCH_PLAN successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration x_detector_pipeline_launchplan type LAUNCH_PLAN successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration y_detector_pipeline_launchplan type LAUNCH_PLAN successful with version Mxuji6SmjvFRo5-uGYS5yg
[✔] Registration z_detector_pipeline_launchplan type LAUNCH_PLAN successful with version Mxuji6SmjvFRo5-uGYS5yg```
z_detector, the bottom one, is the new launch plan that I added
In the Launch Plans tab, I can see the versions of that launchplan showing up, as shown in the first image
then the second image shows that on the Workflows tab, there are no new versions showing up, the latest is from over a month ago
then when I click Launch Workflow, the new launchplan does not show up, because it only has those old workflow versions where it didn't exist. and I also can't run the updated versions of the existing launchplans
ytong
that will let us track down the api call
ytong
and mind screenshotting the ui that is missing?
ytong
not entirely following… would you mind copy/pasting the register output (redact names if you want)
andrew.merrill
That's a bit before I updated my flyte deployment through terraform, I wonder if something broke with that somehow
andrew.merrill
I've ran it a bunch recently, and the most recent workflow version is 8/22/2024, I'm now seeing...
andrew.merrill
Yeah, so all the Registration ... LAUNCH_PLAN
and Registration ... WORKFLOW
stuff prints out. The workflow is there for the overall workflow, and each launchplan that I'd expect to see if there, but then that version of workflow doesn't show up in the console, but it does show up in the launchplans
ytong
and yes the versions should be the same.
ytong
does it show up there?
ytong
the register command should print out all the entities it’s registering
andrew.merrill
Hey! I'm having an issue with registering launchplans, and having them show up connected to a workflow. I'm adding a new launchplan for a workflow using LaunchPlan.get_or_create(...)
, then running pyflyte register path/to/workflow.py -p project -d domain
. Everything runs fine, and it shows the new version string. In the flyte console, I can see the launchplan itself in the Launch Plans tab with the new version, but when I go to the workflow in the Workflows tab, and click Launch Workflow, its only showing older versions in the Workflow version dropdown, and the new launchplan isn't shown in the Launch Plan dropdown.
Is the workflow version the same as the version shown when registering? This only just now seems to be an issue, it was working fine up until recently