zig compatibility notes#

comparison of zig implementation with python prefect server.

RetryFailedFlows#

matching behavior ✓#

aspect	python	zig
from_states	{RUNNING}	StateTypeSet.init(&.{.RUNNING})
to_states	{FAILED}	StateTypeSet.init(&.{.FAILED})
retry check	`retries is None or run_count > retries`	`ctx.retries orelse return; if (ctx.run_count > max_retries) return;`
scheduled time	`now + timedelta(seconds=retry_delay or 0)`	`now_us + delay_seconds * 1_000_000`
retry state	`AwaitingRetry` (SCHEDULED)	`SCHEDULED` with name "AwaitingRetry"
response	`reject_transition(state, reason="Retrying")`	`REJECT` status

implemented ✓#

behavior	status
retry_type = "in_process" on retry	✓ via `set_retry_type_in_process` flag
resuming = false on retry	✓ via `set_resuming_false` flag
pause_keys = [] on retry	✓ via `clear_pause_keys` flag
retry_type = null when exhausted	✓ via `clear_retry_type` flag

not needed#

old client compatibility (API < 0.8.3): python resets failed task runs to AwaitingRetry state - not needed for new clients

schema compatibility#

empirical_policy JSON#

python schema:

{
  "retries": int | null,
  "retry_delay": int | null,
  "pause_keys": [],
  "resuming": false,
  "retry_type": null | "in_process" | "reschedule",
  "max_retries": 0,        // deprecated
  "retry_delay_seconds": 0  // deprecated
}

zig reads: retries, retry_delay from this JSON zig stores: as TEXT (sqlite) or JSONB (postgres) - compatible

storage types#

column	postgres python	postgres zig	sqlite python	sqlite zig
id	UUID	UUID	CHAR(36)	CHAR(36)
timestamps	TIMESTAMP(tz)	TIMESTAMP(tz)	DATETIME	DATETIME
empirical_policy	JSONB	JSONB	JSON	TEXT

migration path#

our zig server stores data compatible with python prefect server:

UUIDs as standard format
timestamps as ISO 8601 UTC
JSON/JSONB for policy fields

data created by zig server can be read by python server (and vice versa).