lei.chen/myxce
1
0
Fork 1
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
myxce/.kiro/specs/worker-join-refactor/tasks.md
Rongsong Shen 93ed282ef7 feat(worker-join-refactor): implement apply_join_state, init_worker, rejoin_worker 
Replace monolithic add_worker with three focused functions:
- apply_join_state (static under production, exported under XCE_TESTING):
  shared status/resource update called by both create and rejoin paths;
  clears XCE_WORKER_UNAVAIL, applies hconfig overrides, updates resources
- init_worker: first-time join path via makeWorker + apply_join_state
- rejoin_worker: reconnect path with explicit admin-bit preservation
  (XCE_WORKER_ADMIN_MASK = LOCKEDU|CLOSED|HIDDEN|CLOSED_BY_ADMIN)

Also makes find_matched_hostconfig non-static so join.c can call it
via the extern declaration already in worker.h.

Removes add_worker (no longer declared in worker.h, no callers).

All 7 tests in test_worker_join pass:
  apply_join_state (clears UNAVAIL, hosttype override, resource values),
  init_worker (creates entry, returns correct idx),
  rejoin_worker (updates attrs/ep, preserves admin bits).
2026-05-20 08:24:26 +00:00

6.5 KiB
Raw Blame History

Implementation Plan

Task Format

  • Major task number and description
  • Sub-task number and description
    • Detail items
    • Observable completion condition
    • Requirements: X.X, Y.Y

  • 1. Foundation: write failing tests for all new worker join behaviors (TDD Red)

  • 1.1 Write failing tests for the shared status and resource update step

    • Add a new test file (e.g., test_worker_join.c) to src/controller/testsuite/ and register it in controller_test.c
    • Tests assert: after the helper runs, XCE_WORKER_UNAVAIL is cleared on the worker's status, hosttype/model match the hconfig override values, slot count equals the hconfig mxj value, and host-attribute resources (type, model, cpuf, maxcpus, maxmem, maxswap, maxtmp, ndisks) hold the expected values
    • Tests compile but fail because the helper does not yet exist
    • Requirements: 4.1, 4.2, 4.3

  • 1.2 Write failing tests for the first-time worker creation path

    • Tests assert: calling the creation function with a previously-unknown hostname produces a non-negative index; key_find on workermgr with that name succeeds after the call
    • Tests assert: when the creation function returns an error, the error code is negative and no on_worker_join side effect occurs
    • Tests compile but fail because the function does not yet exist
    • Requirements: 2.1, 2.2, 2.3

  • 1.3 Write failing tests for the rejoin update path with admin-bit preservation

    • Tests assert: after the update function runs, the worker's host.attrs reflects the new values and w->ep matches the new endpoint
    • Tests assert: admin status bits (XCE_WORKER_LOCKEDU, XCE_WORKER_CLOSED, XCE_WORKER_HIDDEN, XCE_WORKER_CLOSED_BY_ADMIN) present before the call are still set after it
    • Tests compile but fail because the function does not yet exist
    • Requirements: 3.1, 3.2, 3.3
  • 2. Core: extract shared status and resource update into a standalone helper (TDD Green — 1.1)
    • Move the status-clear + hconfig-override + resource-update block out of add_worker in worker.c into a new static helper function
    • The helper covers: clearing XCE_WORKER_UNAVAIL, assigning w->ep, overriding hosttype/model strings from hconfig (with dereference/makeString), calling add_hostattrs_resvals, computing mxj slots via set_host_slots, replacing dispatch_window_config and thresholds from hconfig, and looping over hconfig->nresources with mark_host_res
    • Must be called under the caller-held per-worker write lock; document this precondition with a comment
    • Tests from 1.1 turn green; add_worker still compiles by delegating to the new helper
    • Requirements: 4.1, 4.2, 4.3

  • 3. Core: implement the first-time creation and rejoin update helpers

  • 3.1 Implement the first-time worker creation function (TDD Green — 1.2)

    • Build a temporary xce_worker_t from the incoming hostname and attrs, then call makeWorker to register it in workermgr (makeWorker manages its own write lock)
    • On a NULL return from makeWorker, log the failure and return -errno to the caller without touching on_worker_join
    • On success, acquire the per-worker write lock, call the shared helper from Task 2, release the lock, and return w->idx
    • Tests from 1.2 turn green
    • Requirements: 2.1, 2.2, 2.3

  • 3.2 Implement the rejoin worker update function (TDD Green — 1.3)

    • Acquire the per-worker write lock at the start; save admin_status = w->host.status & XCE_WORKER_ADMIN_MASK before any modification
    • Free and replace the existing hosttype and hostmodel strings; overwrite numeric attrs fields (cpufactor, maxcpus, maxmem, maxswap, maxtmp, ndisks) from the incoming attrs argument
    • Call the shared helper from Task 2 (status/resource update, ep assignment), then restore admin bits via w->host.status |= admin_status; release the lock
    • Return w->idx; tests from 1.3 turn green and admin bits are unchanged after a full round-trip
    • Requirements: 3.1, 3.2, 3.3

  • 4. Integration: wire the explicit existence check and new helpers into the join handler

  • 4.1 Update exported worker function declarations

    • Remove the add_worker declaration from src/controller/include/worker.h
    • Add declarations for the first-time creation function and the rejoin update function with their full parameter lists (including the xce_hostconfig_t* parameter)
    • Verify that the only direct caller of add_worker was handle_worker_join; if any other callers exist, update them to use the appropriate new function
    • worker.h compiles cleanly with no remaining reference to add_worker
    • Requirements: 5.4

  • 4.2 Rewrite the worker join handler with explicit existence check and branching

    • Remove the outer lock(c->workermgr, RTTI_LOCK_WRITE) / unlock pair that currently wraps the add_worker call (eliminates the double write-lock deadlock)
    • Move find_matched_hostconfig to the top of the handler, before any lock; reject the join with the appropriate error code and set reply.status if it returns NULL
    • Acquire a brief workermgr read lock, call key_find, release the read lock immediately
    • On -ENOENT call the first-time creation function; on a found entry call the rejoin update function; on any other error from key_find reject the join and return that error code
    • Fire BEFORE_EVENT / HOOK_DETAIL_WORKER_JOIN before the key_find step; fire AFTER_EVENT after the create-or-update completes; call xce_reply and on_worker_join outside any lock
    • Set priv->u.idx before calling on_worker_join; if create or update returned an error, set reply.status, call xce_reply, and return without calling on_worker_join
    • The join handler compiles and the sequence check → create-or-update → status/resource-update is visible at the call site with no hidden branching inside makeWorker
    • Requirements: 1.1, 1.2, 1.3, 2.2, 2.3, 4.4, 5.1, 5.2, 5.3
  • 5. Validation: run full test suite and confirm behavioral preservation
    • Build the controller test binary and run all suites in src/controller/testsuite/
    • Confirm test_e2e_hooks records BEFORE_EVENT before any workermgr mutation and AFTER_EVENT after it, matching the pre-refactor ordering
    • If any test fixture in the suite constructed workers by calling add_worker directly, update the fixture to call the creation function with the added xce_hostconfig_t* argument
    • make test exits 0; no previously-passing test is broken by the refactor
    • Requirements: 5.1, 5.2, 5.3, 5.4