Duplicate runs scheduled

Description

Occasionally a specific flow will get double-scheduled in the latest version of Prefect server (we’ve run into this bug several times on previous versions). I.e. for a single clock we will get two identical runs at the same time (both visible in the UI, then both executing at the same time).

Unfortunately this does not happen on every scheduler cycle or we could troubleshoot more easily.

This may or may not be relevant, but it only happens on the times when the schedule is set to run with specific parameters. Another clock on the same schedule does not have this problem.

We can temporarily fix it by:

• Toggling the schedule on and off in the UI (only works sometimes).
• Deleting the postgres data and re-registering the flow

Expected Behavior

Any unique combination of time, flow, and parameters should only get scheduled once by the server.

Reproduction

Unfortunately we can’t always reproduce it, but this problem eventually recurs even when we completely wipe the server setup and redeploy.

The schedule in question is generated as:

# normally imported from a config:
CRON_TIMES_LOADONLY = ["0 6 * * *"] 
CRON_TIMES_FULLRUN = ["0 9 * * *"]

start_date = pendulum.now(tz="America/Chicago")  # pin clock to CDT
clocks_loadonly = [
    clocks.CronClock(s, start_date=start_date) for s in CRON_TIMES_LOADONLY
]
clocks_fullrun = [
    clocks.CronClock(
        s, start_date=start_date, parameter_defaults={"run-dbtbuild": True}
    )
    for s in CRON_TIMES_FULLRUN
]
schedule = Schedule(clocks=clocks_loadonly + clocks_fullrun)

This definition style hasn’t caused us any issues when not used with parameter defaults.

Environment

{ “config_overrides”: { “server”: { “host”: true, “telemetry”: { “enabled”: true }, “ui”: { “apollo_url”: true } } }, “env_vars”: [], “system_information”: { “platform”: “Linux-4.15.0-118-generic-x86_64-with-glibc2.10”, “prefect_backend”: “cloud”, “prefect_version”: “0.13.14”, “python_version”: “3.8.5” } }

[sp1thas]

Same here. I also use parameter_defaults. Flow re-registration seems to fix this issue temporarily. In my case, the scheduling is de-regulated after one or two days. Really difficult to reproduce it locally 😕

Environment

{
  "config_overrides": {
    "api": {
      "url": true
    },
    "context": {
      "secrets": false
    },
    "server": {
      "database": {
        "host_port": true
      },
      "ui": {
        "graphql_url": true
      }
    }
  },
  "env_vars": [
    "PREFECT__CONTEXT__SECRETS__...",
    "PREFECT__CONTEXT__SECRETS__..."
  ],
  "system_information": {
    "platform": "Linux-4.19.0-10-amd64-x86_64-with-debian-10.6",
    "prefect_backend": "server",
    "prefect_version": "0.13.14",
    "python_version": "3.7.3"
  }
}

[joshmeek]

Update: the fix for the duplicate scheduled runs has been merged into server and the next core release will use the updated images 👍