Skip to main content

Scale to multiple instances

One instance runs one flow at a time. Nothing inside a flow makes anything concurrent — parallelism in this product means more instances, each with its own queue. (Why.)

1 · Budget the host first

Each instance costs 2 CPU cores and 2 GB RAM at the recommended baseline. Four instances is roughly 8 cores and 8 GB before Windows takes its share. The most common cause of flaky automation at scale is not the flows — it is a host with no headroom.

2 · Clone, don't reconfigure

Use MuMu's Copy (shown here) so every instance keeps the identical 1280×720 · DPI 240 · 30 FPS baseline. Identical resolution is what lets one set of templates serve the whole pool — a mismatched instance flags every queue entry and breaks pixel-based matching.

Then add each one on the Instance Dashboard.

3 · One flow, many queues

Do not duplicate a flow per instance. A queue entry stores its own input values, so the same saved flow serves the whole pool with different parameters:

InstanceFlowrounds
mumu-0Farm Resources3
mumu-1Farm Resources10

Declare the knobs as Flow Inputs, set the values on each queue entry. One flow to maintain.

4 · Stagger the schedules

Every schedule targets one instance. Two things bite at scale:

  • Same instance, several interval schedules — they skip each other (busy = skipped, no catch-up). Stagger the times or use Priority.
  • All instances starting at 08:00 — a host-wide CPU spike. Offset them a few minutes apart.

See Scheduler.

5 · Stop instances repeating each other's work

By default, instances know nothing about each other — memory scopes run, session and persist are per instance. That isolation is usually what you want.

When it is not — ten instances must not all collect the same shared target — write a Runtime Memory entry in the shared scope: it is the one deliberate cross-instance store. The cooldown pattern works across the pool exactly as it does on one instance.

Common mistakes

MistakeResultFix
Expecting one instance to run two queued flows at onceThe queue is a sequenceMore instances
Cloning by hand with a different resolutionTemplates break on that instance onlyMuMu Copy, keep the baseline
One flow copy per instanceN copies to maintainFlow Inputs + per-entry values
Interval schedules colliding on one instanceEndless Skipped (busy)Stagger, or Priority
Using shared scope for everythingInstances block each other's unrelated workshared only for genuinely shared state

Instance Dashboard · Flow Queue · Run unattended