perf(mdm): Route all sync/push actions through Dagster to prevent request hangs

[Copilot]

Synchronous MDM API calls in the request/response cycle (sync fleet, push device config, push policy child devices) block for orgs with many devices. Dagster was already wired up for device CSV imports in apps/mdm/ but five other call sites remained synchronous.

New Dagster infrastructure

• sync_and_push_mdm_devices asset — accepts a fleet_pks: list[int] config and calls sync_fleet() for each specified fleet with push_config=True
• sync_fleets_job — new job targeting the asset, used for the devices-list sync button

Updated call sites

All five sites now follow the same pattern: trigger Dagster when dagster_enabled(), fall back to synchronous behavior when it isn't (local dev / no DAGSTER_URL).

Site	Async action
devices_list "Sync" button	trigger_dagster_job("sync_fleets_job", {fleet_pks: [...]}) with the current org's fleet PKs → "Sync queued" message
device_update_app_user HTMX	trigger_dagster_job("mdm_job", {device_pks: [pk]})
publish_mdm/import_export.py after_import	trigger_dagster_job("mdm_job", {device_pks: [...]})
mdm/views.py _push_policy_to_mdm	Base policy push stays synchronous; child-device pushes queued via mdm_job
mdm/admin.py PolicyAdmin.save_related	Same split: base policy sync, per-device child policies queued

The sync_fleets_job is scoped to the organization whose sync button was pressed — only that org's fleet PKs are passed in the run config, matching the non-Dagster per-fleet loop behaviour.

# Pattern used consistently across all sites
if dagster_enabled():
    run_config = {"ops": {"push_mdm_device_config": {"config": {"device_pks": device_pks}}}}
    trigger_dagster_job(job_name="mdm_job", run_config=run_config)
else:
    for device in child_devices:
        active_mdm.push_device_config(device)

Tests

New test classes added for each site asserting:

• Dagster path: job triggered with correct PKs (fleet or device), no direct MDM calls
• Fallback path: synchronous MDM calls made, Dagster not triggered

[simonkagwi]

@copilot The sync_fleets_job Dagster job should be able to sync specific fleets, instead of always syncing all fleets. When triggered from the devices list page, it should only sync fleets related to the current organization, similar to the non-Dagster flow

[Copilot]

@copilot The sync_fleets_job Dagster job should be able to sync specific fleets, instead of always syncing all fleets. When triggered from the devices list page, it should only sync fleets related to ...

Done in eb27cc1. The sync_and_push_mdm_devices asset now accepts a fleet_pks: list[int] config and iterates through only those specific fleets using sync_fleet() — matching the non-Dagster per-fleet loop. When triggered from the devices list page, the view now passes request.organization.fleets.values_list("pk", flat=True) in the run config so only the org's fleets are synced.