[KV Offload] Rename SecondaryTierManager.get_finished() to get_finished_jobs() (#43870)

Signed-off-by: Ronen Schaffer <ronen.schaffer@ibm.com>
This commit is contained in:
Ronen Schaffer
2026-05-28 16:00:18 +00:00
committed by GitHub
parent 5d126dd155
commit 4bfa0f2b14
6 changed files with 15 additions and 15 deletions
+2 -2
View File
@@ -72,14 +72,14 @@ def make_job(
def drain(tier: FileSystemTierManager, max_rounds: int = 40) -> list: def drain(tier: FileSystemTierManager, max_rounds: int = 40) -> list:
""" """
Call get_finished() repeatedly until no new results arrive for 5 Call get_finished_jobs() repeatedly until no new results arrive for 5
consecutive rounds or max_rounds is reached. consecutive rounds or max_rounds is reached.
""" """
results = [] results = []
idle = 0 idle = 0
for _ in range(max_rounds): for _ in range(max_rounds):
time.sleep(0.01) time.sleep(0.01)
new = list(tier.get_finished()) new = list(tier.get_finished_jobs())
results.extend(new) results.extend(new)
if new: if new:
idle = 0 idle = 0
@@ -184,7 +184,7 @@ class TestTieringOffloadingManager:
# End of step 1: _maybe_process_finished_jobs() was already called by # End of step 1: _maybe_process_finished_jobs() was already called by
# prepare_store() above (setting the per-step flag), so take_events() # prepare_store() above (setting the per-step flag), so take_events()
# does NOT poll get_finished() again — cascade completions remain # does NOT poll get_finished_jobs() again — cascade completions remain
# unprocessed until the next step. # unprocessed until the next step.
list(self.manager.take_events()) list(self.manager.take_events())
+4 -4
View File
@@ -50,7 +50,7 @@ class SecondaryTierManager(ABC):
IMPORTANT: All methods run in the Scheduler process and must be IMPORTANT: All methods run in the Scheduler process and must be
lightweight and non-blocking. submit_load() and submit_store() submit lightweight and non-blocking. submit_load() and submit_store() submit
async jobs; get_finished() polls for completion. async jobs; get_finished_jobs() polls for completion.
""" """
def __init__( def __init__(
@@ -106,7 +106,7 @@ class SecondaryTierManager(ABC):
3. Allocating space in this tier 3. Allocating space in this tier
4. Submitting the async transfer (read from primary via block_ids) 4. Submitting the async transfer (read from primary via block_ids)
Report completion via ``get_finished()``. Report completion via ``get_finished_jobs()``.
Args: Args:
job_metadata: Job metadata including job_id, keys, and block_ids job_metadata: Job metadata including job_id, keys, and block_ids
@@ -131,7 +131,7 @@ class SecondaryTierManager(ABC):
The implementation must copy data from this tier into the The implementation must copy data from this tier into the
primary-tier slots identified by ``block_ids``. primary-tier slots identified by ``block_ids``.
Report completion via ``get_finished()``. Report completion via ``get_finished_jobs()``.
Args: Args:
job_metadata: Job metadata including job_id, keys, and block_ids job_metadata: Job metadata including job_id, keys, and block_ids
@@ -140,7 +140,7 @@ class SecondaryTierManager(ABC):
pass pass
@abstractmethod @abstractmethod
def get_finished(self) -> Iterable[JobResult]: def get_finished_jobs(self) -> Iterable[JobResult]:
""" """
Return all jobs (loads and stores) that completed since the last call. Return all jobs (loads and stores) that completed since the last call.
@@ -61,7 +61,7 @@ class ExampleSecondaryTierManager(SecondaryTierManager):
# key -> True (only care about presence) # key -> True (only care about presence)
self.blocks: dict[OffloadKey, bool] = {} self.blocks: dict[OffloadKey, bool] = {}
# Completed jobs waiting to be retrieved by get_finished() # Completed jobs waiting to be retrieved by get_finished_jobs()
self.completed_jobs: list[JobResult] = [] self.completed_jobs: list[JobResult] = []
def lookup(self, key: OffloadKey, req_context: ReqContext) -> bool | None: def lookup(self, key: OffloadKey, req_context: ReqContext) -> bool | None:
@@ -120,7 +120,7 @@ class ExampleSecondaryTierManager(SecondaryTierManager):
self.completed_jobs.append(JobResult(job_id=job_metadata.job_id, success=True)) self.completed_jobs.append(JobResult(job_id=job_metadata.job_id, success=True))
def get_finished(self) -> Iterable[JobResult]: def get_finished_jobs(self) -> Iterable[JobResult]:
""" """
Poll for finished jobs. Poll for finished jobs.
+2 -2
View File
@@ -47,7 +47,7 @@ class FileSystemTierManager(SecondaryTierManager):
queue, so neither starves. queue, so neither starves.
submit_store / submit_load are non-blocking: they enqueue tasks and return. submit_store / submit_load are non-blocking: they enqueue tasks and return.
get_finished() polls job completion and returns completed JobResults. get_finished_jobs() polls job completion and returns completed JobResults.
""" """
@@ -131,7 +131,7 @@ class FileSystemTierManager(SecondaryTierManager):
) )
self._pool.enqueue_load(job_metadata.job_id, len(job_metadata.keys), tasks) self._pool.enqueue_load(job_metadata.job_id, len(job_metadata.keys), tasks)
def get_finished(self) -> Iterable[JobResult]: def get_finished_jobs(self) -> Iterable[JobResult]:
""" """
Collect completed jobs from the finished-jobs queue. Collect completed jobs from the finished-jobs queue.
""" """
+4 -4
View File
@@ -113,7 +113,7 @@ class TieringOffloadingManager(OffloadingManager):
Key internal state: Key internal state:
- Minimal state tracking; relies on secondary tiers to report completion - Minimal state tracking; relies on secondary tiers to report completion
via get_finished() via get_finished_jobs()
- Secondary tiers return JobResult objects containing all necessary - Secondary tiers return JobResult objects containing all necessary
information information
- job_id_counter: monotonically increasing counter for job IDs - job_id_counter: monotonically increasing counter for job IDs
@@ -182,14 +182,14 @@ class TieringOffloadingManager(OffloadingManager):
Unconditionally poll all secondary tiers for completed jobs. Unconditionally poll all secondary tiers for completed jobs.
This method: This method:
1. Calls get_finished() on each secondary tier 1. Calls get_finished_jobs() on each secondary tier
2. For completed stores (primary→secondary): calls primary.complete_read() 2. For completed stores (primary→secondary): calls primary.complete_read()
to decrement ref_cnt to decrement ref_cnt
3. For completed loads (secondary→primary): calls primary.complete_write() 3. For completed loads (secondary→primary): calls primary.complete_write()
to make blocks available to make blocks available
""" """
for i, tier in enumerate(self.secondary_tiers): for i, tier in enumerate(self.secondary_tiers):
for completed_job in tier.get_finished(): for completed_job in tier.get_finished_jobs():
job_id = completed_job.job_id job_id = completed_job.job_id
job_metadata = self._transfer_jobs.pop(job_id, None) job_metadata = self._transfer_jobs.pop(job_id, None)
assert job_metadata is not None, ( assert job_metadata is not None, (
@@ -464,7 +464,7 @@ class TieringOffloadingManager(OffloadingManager):
tier.submit_store(job_metadata) tier.submit_store(job_metadata)
# Note: The async transfers are now in flight. Their completion is # Note: The async transfers are now in flight. Their completion is
# tracked via get_finished() / _maybe_process_finished_jobs(). # tracked via get_finished_jobs() / _maybe_process_finished_jobs().
def take_events(self) -> Iterable[OffloadingEvent]: def take_events(self) -> Iterable[OffloadingEvent]:
""" """