• j

    Jonas Hanfland

    1 year ago
    Hey guys, I am getting an exception when trying to run a
    BigQueryTask
    . This is the task I am trying to run:
    BigQueryTask(
        name="Get all statement IDs",
        query="""
            SELECT * FROM temporary_tables.statements
        """,
        max_retries=3,
        retry_delay=timedelta(seconds=30),
        to_dataframe=True,
    )
    and it throws this exception: (see thread) All the times this task has been run in the past it passed just fine. Does anybody have an idea what might be the issue? Thanks very much in advance!
    j
    Chris White
    6 replies
    Copy to Clipboard
  • ale

    ale

    1 year ago
    Hey folks, anyone experiences issues when creating task definitions on Fargate. I mean, task definitions are created, but at task setup I get the following error
    ale
    Chris White
    +2
    52 replies
    Copy to Clipboard
  • Krzysztof Nawara

    Krzysztof Nawara

    1 year ago
    Hey, nice to meet everyone 🙂 I've got question about caching in Prefect - I'm trying to build a ML pipeline which can take advantage of caching, but can also selectively recompute some of the nodes (e.g. because logic changed, or new node was added). For that I was planning to keep track of version for each task and make it part of the cache key/target location. The bit I'm struggling the most with is invalidation of downstream tasks - if the task version get's bumped, downstreams cannot be read from cache. So far I've looked at cache_key and cache_validator. If validators could access states of current execution of upstream tasks, they would be able to return False is state of at least one of the upstreams is different from Cached. Really neat solution, but from what I'm seeing cache_validator only get's access to previous state of current tasks and values of inputs, but not their states. I also looked at Result and target, but they seem even more restrictive, at least until PIN 16 is fully implemented. The only workaround I can think of is making version of each upstream task part of the cache key for the task. This is going to get tedious real fast unless there is a way to propagate that version through the pipeline and generate the cache keys/targets. But this feels a lot like a hack, so before attempting that I wanted to ask here if experts here can think of a better options. Thank you very much in advance, Chris
    Krzysztof Nawara
    Jim Crist-Harif
    +1
    15 replies
    Copy to Clipboard
  • d

    Dolor Oculus

    1 year ago
    I feel like I've seen this in the docs, but I can't find it now 😕 If I want to configure which server my code uses (ie dev vs uat vs prod), I am currently doing it by setting
    PREFECT__SERVER__ENDPOINT
    (in bash, before invoking Python). Are their other ways of configuring this? (ie via constructor arguments at runtime)? ty!
    d
    1 replies
    Copy to Clipboard
  • m

    Marwan Sarieddine

    1 year ago
    Hi folks, we occasionally face this error when running our flow on EKS with a static Dask cluster setup (i.e. using a LocalEnvironment with a DaskExecutor) and a kubernetes agent: (please see the thread for more details)
    m
    Jim Crist-Harif
    +1
    6 replies
    Copy to Clipboard
  • Mike Fransesco

    Mike Fransesco

    1 year ago
    Hello 1st message! Our company is looking at prefect and I am enquiring if there is any active support for Docker Swarm vs Kuberneties ?
    Mike Fransesco
    Jim Crist-Harif
    13 replies
    Copy to Clipboard
  • Chirag

    Chirag

    1 year ago
    Hi Everyone, I have a doubt regarding the metadata storage of task results. Is there a way to delete the task metadata once the flow is completed? Currently, I am storing the task results in S3.
    Chirag
    j
    2 replies
    Copy to Clipboard
  • s

    Sven Teresniak

    1 year ago
    i'm using a small dask cluster, server flavor, v0.13.9, localagent. After upgrade to v0.13.10 I cannot start a flow. The agent prints
    [2020-10-07 14:26:37,427] ERROR - agent1 | 400 Client Error: Bad Request for url: <http://localhost:4200/>
    But it is still possible to register flows as usual. I seem them in the UI, too. I switched back to v0.13.9 and the error is gone. I only changed the Prefect version -- no configuration and no flow code. I'd like to provide you with more information about this bug but the agent is ignoring
    $PREFECT__LOGGING__LEVEL=DEBUG
    it seems. Is there any change in the enviroment for the agent? Why can I register flows but not start them?
    s
    j
    +1
    21 replies
    Copy to Clipboard
  • Matias Godoy

    Matias Godoy

    1 year ago
    Hi guys! I wanted to ask a question about flow labels; Right now I have two environments (DEV and PROD) and I have 2 different Kubernetes Agents (one with label
    dev
    and the other with label
    prod
    ), which also means that I'm using Docker Storage for the flows. Inside my flow I set the labels like this:
    flow.environment = LocalEnvironment(labels=['dev'])
    The thing is that I'd like the flow label to be "dynamic" in the sense that if I register that flow from a development environment, the label is automatically set to
    dev
    , but if I want to register it for production, I'd like the flow label to be
    prod
    . For now I register the flows from my own laptop, which automatically generates the Docker image and uploads it to my ECR for the Agents to use. Maybe I'm wrong with this, but I guess that if I set an environment variable in my local computer, and make the flow code something like
    flow.environment = LocalEnvironment(labels=[my_env_var])
    it will not work, because that variable will not exist in the container that runs the flow. Is this correct? Another alternative would be to be able to set flow labels when registering the flow using the CLI. Something like
    prefect register flow --file my_flow.py --name My-Flow --label dev
    , but according to the CLI help,
    --label
    is not a valid parameter when registering flows. Do you have any recommendations for setting flow labels dynamically depending on the environment they where registered? Thanks!
    Matias Godoy
    j
    3 replies
    Copy to Clipboard
  • Thomas La Piana

    Thomas La Piana

    1 year ago
    does prefect have the "sensor" concept from airflow? triggering a flow based on a change in an external object's state?
    Thomas La Piana
    Jeremiah
    2 replies
    Copy to Clipboard