perf(search): cache metric instruments in hot vector-sync path

Codex review on PR #754 flagged that _log_vector_sync_complete runs
per entity in large sync batches, and each call now goes through
logfire.metric_histogram(...).record(...) — OTel's MeterProvider
lookup is O(1) but still non-free under load. The original
telemetry wrapper cached instruments in a process-wide dict; my
strip lost that.

Restores the cache as two small module-level helpers in
search_repository_base.py (_metric_histogram / _metric_counter)
so instruments are resolved once and reused. Callers switch to
the helpers (11 sites).

Tests clear _METRIC_INSTRUMENTS via monkeypatch so fake factories
win on the first lookup.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>

Author: phernandez

src/basic_memory/repository/search_repository_base.py

@@ -41,6 +41,29 @@
 _SQLITE_MAX_PREPARE_WINDOW = 8
 
 
+# Cache instruments so the per-entity hot path in _log_vector_sync_complete
+# doesn't re-enter the OTel MeterProvider lookup on every sample.
+_METRIC_INSTRUMENTS: dict[tuple[str, str, str], Any] = {}
+
+
+def _metric_histogram(name: str, unit: str = "") -> Any:
+    key = ("histogram", name, unit)
+    instrument = _METRIC_INSTRUMENTS.get(key)
+    if instrument is None:
+        instrument = logfire.metric_histogram(name, unit=unit)
+        _METRIC_INSTRUMENTS[key] = instrument
+    return instrument
+
+
+def _metric_counter(name: str) -> Any:
+    key = ("counter", name, "")
+    instrument = _METRIC_INSTRUMENTS.get(key)
+    if instrument is None:
+        instrument = logfire.metric_counter(name)
+        _METRIC_INSTRUMENTS[key] = instrument
+    return instrument
+
+
 @dataclass
 class VectorSyncBatchResult:
     """Aggregate result for batched semantic vector sync runs."""
@@ -1069,7 +1092,7 @@ def emit_progress(entity_id: int) -> None:
                 write_seconds_total=result.write_seconds_total,
             )
             batch_total_seconds = time.perf_counter() - batch_start
-            logfire.metric_histogram(
+            _metric_histogram(
                 "vector_sync_batch_total_seconds",
                 unit="s",
             ).record(
@@ -1079,42 +1102,42 @@ def emit_progress(entity_id: int) -> None:
                     "skip_only_batch": result.embedding_jobs_total == 0,
                 },
             )
-            logfire.metric_counter("vector_sync_entities_total").add(
+            _metric_counter("vector_sync_entities_total").add(
                 result.entities_total,
                 attributes={
                     "backend": backend_name,
                     "skip_only_batch": result.embedding_jobs_total == 0,
                 },
             )
-            logfire.metric_counter("vector_sync_entities_skipped").add(
+            _metric_counter("vector_sync_entities_skipped").add(
                 result.entities_skipped,
                 attributes={
                     "backend": backend_name,
                     "skip_only_batch": result.embedding_jobs_total == 0,
                 },
             )
-            logfire.metric_counter("vector_sync_entities_deferred").add(
+            _metric_counter("vector_sync_entities_deferred").add(
                 result.entities_deferred,
                 attributes={
                     "backend": backend_name,
                     "skip_only_batch": result.embedding_jobs_total == 0,
                 },
             )
-            logfire.metric_counter("vector_sync_embedding_jobs_total").add(
+            _metric_counter("vector_sync_embedding_jobs_total").add(
                 result.embedding_jobs_total,
                 attributes={
                     "backend": backend_name,
                     "skip_only_batch": result.embedding_jobs_total == 0,
                 },
             )
-            logfire.metric_counter("vector_sync_chunks_total").add(
+            _metric_counter("vector_sync_chunks_total").add(
                 result.chunks_total,
                 attributes={
                     "backend": backend_name,
                     "skip_only_batch": result.embedding_jobs_total == 0,
                 },
             )
-            logfire.metric_counter("vector_sync_chunks_skipped").add(
+            _metric_counter("vector_sync_chunks_skipped").add(
                 result.chunks_skipped,
                 attributes={
                     "backend": backend_name,
@@ -1694,7 +1717,7 @@ def _log_vector_sync_complete(
     ) -> None:
         """Log completion and slow-entity warnings with a consistent format."""
         backend_name = type(self).__name__.removesuffix("SearchRepository").lower()
-        logfire.metric_histogram(
+        _metric_histogram(
             "vector_sync_prepare_seconds",
             unit="s",
         ).record(
@@ -1704,7 +1727,7 @@ def _log_vector_sync_complete(
                 "skip_only_entity": entity_skipped and embedding_jobs_count == 0,
             },
         )
-        logfire.metric_histogram(
+        _metric_histogram(
             "vector_sync_queue_wait_seconds",
             unit="s",
         ).record(
@@ -1714,7 +1737,7 @@ def _log_vector_sync_complete(
                 "skip_only_entity": entity_skipped and embedding_jobs_count == 0,
             },
         )
-        logfire.metric_histogram(
+        _metric_histogram(
             "vector_sync_embed_seconds",
             unit="s",
         ).record(
@@ -1724,7 +1747,7 @@ def _log_vector_sync_complete(
                 "skip_only_entity": entity_skipped and embedding_jobs_count == 0,
             },
         )
-        logfire.metric_histogram(
+        _metric_histogram(
             "vector_sync_write_seconds",
             unit="s",
         ).record(

tests/repository/test_semantic_search_base.py

@@ -712,6 +712,8 @@ def add(self, amount, attributes=None) -> None:
 
     monkeypatch.setattr(repo, "_prepare_entity_vector_jobs_window", _stub_prepare_window)
     monkeypatch.setattr(repo, "_flush_embedding_jobs", _stub_flush)
+    # Reset the module-level metric cache so the fake factories below win.
+    monkeypatch.setattr(search_repository_base_module, "_METRIC_INSTRUMENTS", {})
     monkeypatch.setattr(
         search_repository_base_module.logfire,
         "metric_histogram",