src/eloq_store.cpp (1)

933-938: Make the global scheduler bounded end-to-end.

Each handler now limits concurrent dispatch, but it still allocates one subrequest object per partition before the first send. That keeps heap growth proportional to partition count and repeats the same scheduler skeleton three times. A small shared helper that keeps only a bounded pool of live subrequests and creates them lazily would remove the duplication and make max_global_request_batch cover memory as well as concurrency.

Also applies to: 940-1008, 1176-1187, 1192-1260, 1398-1407, 1409-1496

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/eloq_store.cpp` around lines 933 - 938, The current handlers eagerly
allocate one subrequest per partition (e.g., creating TruncateRequest instances
and pushing into req->truncate_reqs_) which allows heap growth proportional to
partition count; refactor by introducing a small shared helper (e.g., a
BoundedSubrequestPool or make_lazy_subrequest_generator) that maintains at most
max_global_request_batch live subrequests, creates TruncateRequest (and the
other subrequest types used in the other handlers) lazily when about to
dispatch, and recycles/completes them when a dispatched subrequest finishes;
replace the per-partition eager loops that push into truncate_reqs_ with
iteration that obtains subrequests from this pool and sends them, and reuse the
same helper across the three similar handler sites (the ranges called out) so
memory is bounded by max_global_request_batch as well as concurrency.

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

src/eloq_store.cpp (1)

945-1013: LGTM - Well-structured batch scheduling pattern.

The DropTableScheduleState implementation is sound:

• Proper use of shared_from_this() for safe callback lifetime management
• Correct memory ordering: acq_rel on pending_ ensures visibility of first_error_ writes
• Graceful failure handling with callback clearing before SetDone

Note that DropTableScheduleState, ArchiveScheduleState, and ReopenScheduleState are nearly identical. Consider consolidating into a template or helper to reduce duplication in a follow-up PR.
,

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/eloq_store.cpp` around lines 945 - 1013, The three nearly identical
scheduler structs (DropTableScheduleState, ArchiveScheduleState,
ReopenScheduleState) should be consolidated into a single templated or generic
helper to remove duplication: extract the common members (store, req, total,
next_index) and methods (HandleTruncateResult, OnTruncateDone, ScheduleNext)
into a template class or a policy-based helper that accepts the specific request
type (e.g., TruncateRequest container/accessor) and callback behavior; update
the call sites to instantiate the template with the appropriate request type and
preserve current semantics (shared_from_this usage, memory orderings, error
handling) so functionality and lifetime management remain unchanged.