Files
ClaudeDo/src/ClaudeDo.Worker/Planning/PlanningChainCoordinator.cs
T
mika kuns 816f247d90 fix(worker): keep planning-chain cascade moving past an Idle middle link
OnChildFinishedAsync ignored CancelAsync's result. If a chain successor
sat in a non-cancellable state (e.g. parked to Idle out of band) when
its predecessor failed/was cancelled, CancelAsync was a silent no-op and
the cascade stopped there, leaving the rest of the chain Queued+blocked
forever. Now it walks past any link CancelAsync can't touch and keeps
cancelling downstream.
2026-07-23 17:09:55 +02:00

134 lines
5.7 KiB
C#

using ClaudeDo.Data;
using ClaudeDo.Data.Models;
using ClaudeDo.Worker.State;
using Microsoft.EntityFrameworkCore;
using TaskStatus = ClaudeDo.Data.Models.TaskStatus;
namespace ClaudeDo.Worker.Planning;
public sealed class PlanningChainCoordinator
{
private readonly IDbContextFactory<ClaudeDoDbContext> _dbFactory;
private readonly Func<ITaskStateService> _state;
public PlanningChainCoordinator(
IDbContextFactory<ClaudeDoDbContext> dbFactory,
Func<ITaskStateService> state)
{
_dbFactory = dbFactory;
_state = state;
}
// Sets up a sequential chain over a planning parent's children.
// - First non-terminal child gets BlockedByTaskId=null.
// - Each subsequent non-terminal child gets BlockedByTaskId=<predecessor>,
// so the picker skips them until the predecessor finishes.
// - When enqueue is true, each non-terminal child is also set to Status=Queued
// (the user-driven "Queue plan"). When false (finalize), children are left
// Idle and only the blocked-by links are established, so nothing runs until
// the user queues the plan.
// - Terminal children (Done/Failed/Cancelled) are left untouched; they are
// skipped when computing predecessors so a re-run on a partially executed
// chain leaves history alone but still reshapes the tail.
// - Running children abort the operation — the chain cannot be reshaped while
// one of its members is mid-flight.
// Returns the number of children placed in the chain.
internal async Task<int> SetupChainAsync(string parentTaskId, bool enqueue, CancellationToken ct = default)
{
await using var ctx = await _dbFactory.CreateDbContextAsync(ct);
var parent = await ctx.Tasks.FirstOrDefaultAsync(t => t.Id == parentTaskId, ct)
?? throw new InvalidOperationException($"Task {parentTaskId} not found.");
var children = await ctx.Tasks
.Where(t => t.ParentTaskId == parentTaskId)
.OrderBy(t => t.SortOrder).ThenBy(t => t.CreatedAt)
.ToListAsync(ct);
if (children.Count == 0)
throw new InvalidOperationException("Parent has no subtasks.");
var running = children.FirstOrDefault(c => c.Status == TaskStatus.Running);
if (running is not null)
throw new InvalidOperationException(
$"Child {running.Id} is running; cannot reshape chain.");
// Re-shape over Idle and Queued children only; leave Done/Failed/Cancelled
// (terminal) results in place.
var sequenceable = children
.Where(c => c.Status == TaskStatus.Idle || c.Status == TaskStatus.Queued)
.ToList();
var state = _state();
for (int i = 0; i < sequenceable.Count; i++)
{
if (enqueue)
await state.EnqueueAsync(sequenceable[i].Id, ct);
if (i == 0)
await state.UnblockAsync(sequenceable[i].Id, ct);
else
await state.BlockOnAsync(sequenceable[i].Id, sequenceable[i - 1].Id, ct);
}
return sequenceable.Count;
}
// User-triggered "send plan to queue". Only valid once the plan is finalized
// (children are "Planned"); otherwise the children are still drafts.
public async Task<int> QueuePlanAsync(string parentTaskId, CancellationToken ct = default)
{
await using var ctx = await _dbFactory.CreateDbContextAsync(ct);
var phase = await ctx.Tasks.AsNoTracking()
.Where(t => t.Id == parentTaskId)
.Select(t => (PlanningPhase?)t.PlanningPhase)
.FirstOrDefaultAsync(ct);
if (phase is null)
throw new InvalidOperationException($"Task {parentTaskId} not found.");
if (phase != PlanningPhase.Finalized)
throw new InvalidOperationException("Plan must be finalized before it can be queued.");
return await SetupChainAsync(parentTaskId, enqueue: true, ct);
}
public async Task<string?> OnChildFinishedAsync(
string childTaskId, TaskStatus finalStatus, CancellationToken ct = default)
{
var nextId = await FindSuccessorAsync(childTaskId, ct);
if (nextId is null) return null;
if (finalStatus == TaskStatus.Done)
{
await _state().UnblockAsync(nextId, ct);
return nextId;
}
// Child failed or was cancelled: cancel the immediate successor so the chain is
// not left wedged. If it's cancellable, CancelAsync's own OnChildTerminalAsync
// callback recurses into this method for its successor, cascading cancellation
// through the rest of the chain. If it's not (e.g. it was parked back to Idle
// out of band), CancelAsync is a no-op and nothing will ever call back for it —
// keep walking the chain ourselves so the tail isn't left wedged forever.
var predecessorId = nextId;
while (true)
{
var result = await _state().CancelAsync(predecessorId, DateTime.UtcNow, ct);
if (result.Ok) return null;
var following = await FindSuccessorAsync(predecessorId, ct);
if (following is null) return null;
predecessorId = following;
}
}
// The successor is whichever sibling explicitly blocks on this task.
private async Task<string?> FindSuccessorAsync(string taskId, CancellationToken ct)
{
await using var ctx = await _dbFactory.CreateDbContextAsync(ct);
return await ctx.Tasks
.AsNoTracking()
.Where(t => t.BlockedByTaskId == taskId)
.OrderBy(t => t.SortOrder).ThenBy(t => t.CreatedAt)
.Select(t => t.Id)
.FirstOrDefaultAsync(ct);
}
}