Sure, where can I find this file?

Yes that’s what I did initially, but eventually I decided that it’s too much complexity added for nothing really, I’d rather drop omegaconf and if one day clearml supports it out of the box take advantage of it

How exactly is the clearml-agent killing the task?

yes

The task I cloned from is not the one I though

AgitatedDove14 I see that the default sample_frequency_per_sec=2. , but in the UI, I see that there isn’t such resolution (ie. it logs every ~120 iterations, corresponding to ~30 secs.) What is the difference with report_frequency_sec=30. ?

Nice, thanks!

The rest of the configuration is set with env variables

might be worth documenting 😄

Hi AgitatedDove14 , thanks for the answer! I will try adding 'multiprocessing_context='forkserver' to the DataLoader. In the issue you linked, nirraviv mentionned that forkserver was slower and shared a link to another issue https://github.com/pytorch/pytorch/issues/15849#issuecomment-573921048 where someone implemented a fast variant of the DataLoader to overcome the speed problem.
Did you experiment any drop of performances using forkserver? If yes, did you test the variant suggested i...

Ok interestingly using matplotlib.use('agg') it doesn't leak (idea from here )

here is the function used to create the task:

` def schedule_task(parent_task: Task,
task_type: str = None,
entry_point: str = None,
force_requirements: List[str] = None,
queue_name="default",
working_dir: str = ".",
extra_params=None,
wait_for_status: bool = False,
raise_on_status: Iterable[Task.TaskStatusEnum] = (Task.TaskStatusEnum.failed, Task.Ta...

In execution tab, I see old commit, in logs, I see an empty branch and the old commit

Hi SuccessfulKoala55 , Can the new accounts (password-protected) have the same names?

AgitatedDove14 yes 🙂

The jump in the loss when resuming at iteration 31 is probably another issue -> for now I can conclude that:
I need to set sdk.development.report_use_subprocess = false I need to call task.set_initial_iteration(0)

I also tried task.set_initial_iteration(-task.data.last_iteration) , hoping it would counteract the bug, didn’t work

Yes, I get 0 afterwards

well I still see some ES errors in the logs
` clearml-apiserver | [2021-07-07 14:02:17,009] [9] [ERROR] [clearml.service_repo] Returned 500 for events.add_batch in 65750ms, msg=General data error: err=('500 document(s) failed to index.', [{'index': {'_index': 'events-training_stats_scalar-d1bd92a3b039400cbafc60a7a5b1e52b', '_type': '_doc', '_id': 'c2068648d2fe5da975665985f44c20b6', 'status':..., extra_info=[events-training_stats_scalar-d1bd92a3b039400cbafc60a7a5b1e52b][0] primary shard is not...

but not as much as the ELB reports

Any chance this is reproducible ?

Unfortunately not at the moment, I could find a reproducible scenario. If I clone a task that was stuck and start it, it might not be stuck

How many processes do you see running (i.e. ps -Af | grep python) ?

I will check that when the next one will be blocked 👍

What is the training framework? is it multiprocess ? how are you launching the process itself? is it Linux OS? is it running inside a specific container ?

I train with p...

SuccessfulKoala55 Thanks to that I was able to identify the most expensive experiments. How can I count the number of documents for a specific series? Ie. I suspect that the loss, that is logged every iteration, is responsible for most of the documents logged, and I want to make sure of that

Here I have to do it for each task, is there a way to do it for all tasks at once?

But I would need to reindex everything right? Is that a expensive operation?

I should also rename /opt/trains/data/elastic_migrated_2020-08-11_15-27-05 folder to /opt/trains/data/elastic before running the migration tool right?

The part where I'm lost is why would you need the path to the temp venv the agent creates/uses ?

let's say my task is calling a bash script, and that bash script is calling another python program, I want that last python program to be executed with the environment that was created by the agent for this specific task

AgitatedDove14 I eventually found a different way of achieving what I needed

Is there any logic on the server side that could change the iteration number?

btw I monkey patched ignite’s function global_step_from_engine to print the iteration and passed the modified function to the ClearMLLogger.attach_output_handler(…, global_step_transform=patched_global_step_from_engine(engine)) . It prints the correct iteration number when calling ClearMLLogger.OutputHandler.__ call__ .
` def call(self, engine: Engine, logger: ClearMLLogger, event_name: Union[str, Events]) -> None:

    if not isinstance(logger, ClearMLLogger):
  ...

Ok, I guess I’ll just delete the whole loss series. Thanks!