Early session release: close TM1 sessions when instance has no remaining tasks

[nicolasbisurgi]

Problem Description

When running workflows in --exclusive mode, RushTI holds connections to all TM1 instances for the entire duration of the workflow, even if an instance's tasks finish early.

Example: A workflow with 100 tasks connects to TM1_Instance_A for the first 5 tasks (~10 seconds) and TM1_Instance_B for the remaining 95 tasks (~30 minutes). The current implementation holds the exclusive session on TM1_Instance_A for 30 minutes and 10 seconds, blocking other RushTI instances from using TM1_Instance_A — even though its work completed in 10 seconds.

This applies to both exclusive and non-exclusive modes: idle sessions consume server resources unnecessarily.

Proposed Solution

Implement early session release in the DAG execution loop (work_through_tasks_dag). After each task completes, check whether there are any remaining tasks (pending or running) for the same TM1 instance. If none remain, immediately logout from that instance.

Implementation approach:

1. Build a per-instance task counter at workflow start from all tasks in the DAG
2. Decrement the counter when a task completes
3. When an instance's counter reaches 0, call logout() on that instance's TM1Service, log the early release, and remove it from the active services dict
4. Respect tm1_preserve_connections and force logic (exclusive mode force-closes preserved connections)
5. Skip early release for instances that still have pending/running tasks

Benefits:

• Reduces exclusive mode blocking time significantly for multi-instance workflows
• Frees TM1 server resources (session slots, memory) as early as possible
• No behavior change for single-instance workflows

Alternatives Considered

• No change (status quo): Sessions remain open until workflow ends. Simple but wasteful.
• Periodic session cleanup sweep: Check all instances periodically (e.g., every N tasks) instead of on every completion. Adds complexity without meaningful benefit since the per-completion check is O(1) with a counter.

Additional Context

The change is localized to execution.py (work_through_tasks_dag and logout) and the DAG class (adding a method to query remaining tasks per instance). The logout function already supports selective logout via tm1_preserve_connections and force parameters.