Problemsby GlyphPrefect users keep filing bugs
20 posts from 17 people17d activeFading, +0 this week
Posts per day
The posts behind it
Support deploying flows created by a shared factory via `prefect deploy`
### Describe the current behavior In our use case, the shared Python script already contains the workflow logic. Most business workflows differ only in their input data directory. These commands create two deployments of the same flow. `--name` sets the deployment name, not the...
Support TTL-based caching of external job IDs across task retries
### Describe the current behavior We use Prefect to integrate with an internal compute platform: submit a job for asynchronous execution, obtain its ID, and then poll its status. The platform does not support idempotent submission, so a task retry after a polling error may...
V2 UI feedback: Navigation or usability
### UI switch feedback - From UI: V2 UI - To UI: V1 UI - Current path: /deployments - Reason: Navigation or usability ### Additional notes really bad overview for the upcoming runs, if there are too many scheduled you cant see the recent runs, the deployment overview page should...
`Flow.serialize_parameters` crashes on errors other than `TypeError`/`ValueError`/`RecursionError` from `jsonable_encoder`
### Bug summary `Flow.serialize_parameters` falls back to the `" "` placeholder only for `TypeError`, `ValueError` and `RecursionError`. `jsonable_encoder` deep-copies dataclasses, so any other exception raised by a parameter's `__reduce__` escapes, and the flow run fails before...
V2 UI feedback: Bug or incorrect behavior
### Bug summary ### UI switch feedback - From UI: V2 UI - To UI: V1 UI - Current path: /flows - Reason: Bug or incorrect behavior ### Additional notes When searching for Flows on the Flow, typing a letter in the search bar forces an entire page refresh ### Version info ###...
`prefect automation delete --all` leaves automations behind when there are more than 200
### Bug summary `prefect automation delete --all` reads automations with a single `client.read_automations()` call and no paging. The server returns at most `PREFECT_SERVER_API_DEFAULT_LIMIT` rows (200 by default) for that call, so with more than 200 automations the command: -...
Sending events and API logs from a run uses more CPU than it needs to
### Describe the current behavior A flow run that writes many logs or emits many events spends a lot of CPU in the background services that send them to the API. Against a local `prefect server start`, 20,000 run-logger lines cost 1.43 s of process CPU, and 5,000 `emit_event`...
bug: rrule schedules built from pytz timezones silently lose the timezone (fall back to UTC)
### Bug summary `_iana_timezone_name` (in `prefect.client.schemas.schedules`) does not recognize `pytz` timezones: `pytz` zones expose their IANA name as `.zone`, which is not among the attributes checked (`key`, `name`, `_filename`), so the function falls through to `UTC` on...
ProcessPoolTaskRunner ignores tags() around later submits
### Bug summary `ProcessPoolTaskRunner` serializes the caller's context on the first `submit` and reuses that snapshot for every later submit in the same runner lifecycle. A task submitted inside `with tags("late"):` after an earlier untagged submit runs in the subprocess...
Task parameters that are dataclasses with init=False fields fail to resolve
### Bug summary A task parameter that's a dataclass with a `field(init=False)` fails to resolve once anything inside it changes, for example a future that gets replaced by its result: The same flow works if `retries` is a normal field. `visit_collection` rebuilds a modified...
Concurrency limits page only fetches 200 limits
### Bug summary # Description The Prefect UI `/v2/concurrency-limits` page loads the concurrency limits once, when the page opens, with no limit parameter. The server applies its default page size (PREFECT_SERVER_API_DEFAULT_LIMIT, 200), so the client never receives more than...
Preferred way to cite prefect
### Describe the current behavior No current description on Prefect citation. ### Describe the proposed behavior Somewhere include the preferred Prefect citation. ### Example Use _No response_ ### Additional context _No response_
retry_async_fn(base_delay=0) raises ZeroDivisionError from the backoff helper instead of retrying
### Bug summary `prefect._internal.retries.exponential_backoff_with_jitter()` hands a zero average interval to `clamped_poisson_interval()`, which divides by it: `clamped_poisson_interval()` reaches `exponential_cdf()`, whose first line is `ld = 1 / average_interval`...
Cached task runs wait for a tag concurrency slot before the cache lookup
### Bug summary When a task has a tag with a concurrency limit, the engine takes the tag's slot before it checks the cache. A task run that would come back Cached still waits for a slot behind the live runs holding the limit. Expected: returned Cached after 0.xxs Actual: cache...
Releasing a concurrency slot doesn't wake waiters in the same process, so the slot sits idle until their Retry-After expires
### Bug summary When a caller releases a concurrency slot, nothing tells callers waiting on the same limit in the same process. They keep sleeping until their Retry-After expires and only then ask again, so the slot sits unused in the meantime, even though the process just...
Fix V2 run-list wrapping and toolbar overflow at constrained widths
### Bug summary V2's deployment **Runs** tab does not fit its available content width at a 1280 × 720 viewport. The run count and filters share one line, and the filters extend into the right metadata column. Run-card metadata also breaks within labels such as `0 Parameters` and...
Add a Prefect worker for Nebius Serverless Jobs
### Describe the current behavior Running Prefect flows on Nebius Serverless Jobs currently requires custom integration code to submit containers, monitor jobs, collect logs and propagate cancellation. Users must also handle ambiguous submission responses and reconnect to jobs...
[docs] Improve AutoEnum class documentation in API reference
Currently, all Enum classes in Prefect that inherit from AutoEnum look exactly like this: > **Methods:** > > #### `auto` > > > > Exposes `enum.auto()` to avoid requiring a second import to use `AutoEnum` This makes it difficult for someone to quickly locate the _actual_ enum...
Automatically clean up offline workers
### Describe the current behavior When running multiple Prefect worker replicas on Kubernetes, each pod gets a unique worker name. During rolling deployments, old workers become OFFLINE, but their records stay forever. We can't use a shared stable name because the replicas need...
{{ event.url }} renders empty in automation Jinja templates despite #19851 fix
### Bug summary Summary {{ event.url }} renders as an empty string in automation action Jinja templates on Prefect Cloud. The fix in #19851 (merged Dec 19, 2025) added a url property to the server-side ReceivedEvent, but 8 months later the property still evaluates to empty when...
Companies and products named
About this problem
Evidence
20 posts from 17 people in 1 place, about 2 a week over 66 days. Mostly on PrefectHQ/prefect. Tools named alongside: Google Cloud, Nebius, GitHub, Kubernetes.
Frustration Frustration 0 of 3· Seen on GitHub issues
How it was grouped
Posts about Prefect that state a pain but match no known issue yet. First post Aug 27, 2026, latest Oct 31, 2026. Corroborated: 3 or more posts from 2 or more people or places. Method
Other Prefect problems
Prefect upgrades break working setups
15 posts from 14 people on GitHub issues.
Trouble with Prefect flow
Top post: "Flow run graph polls graph-v2 every second for the whole run, rebuilding the full graph each time", one of 14 posts from 11 people on GitHub issues.
Trouble with Prefect deployment
Top post: "`prefect deployment delete --all` leaves deployments behind when there are more than 200", one of 2 posts from 2 people on GitHub issues. Unverified.
Trouble with Prefect missing
Top post: "V2 UI feedback: Missing feature - Filter by tags search box", one of 2 posts from 2 people on GitHub issues. Unverified.
Prefect rate limits and usage caps block work
2 posts from 2 people on GitHub issues. Unverified.
Trouble with Prefect deployments
One post on GitHub issues: "(ui-v2) Quick Run from deployments list navigates away when selecting from enum dropdown". Unverified.
History