| Long delay before remote scan starts |
Too many folders/files; broad object listing; too many snapshots or manifests; cold catalog metadata. |
Measure planning separately. Improve path pruning, validate strict_column_order, use only Oracle-supported Iceberg metadata maintenance documented for the access path, or prefill the catalog metadata cache. |
| Selective query reads many bytes |
Missing partition predicate, weak file statistics, residual filter, or incompatible data types. |
Inspect predicates and partition information; simplify filters; align types; repair file layout or metadata. |
| Wide query is much slower than narrow query |
Columnar projection is effective but the workload requests too many columns. |
Remove unused columns or create a purpose-built materialized representation for the wide workload. |
| Many tiny remote requests |
Small-file problem. |
For Iceberg, the Oracle-managed compaction will be introduced in the upcoming refresh releases. For non-Iceberg data, reduce file count during data production while preserving useful partition and row-group boundaries. |
| Repeated dashboard queries pay the same remote cost |
No eligible local data cache or ineffective cache policy. |
Use Lake Cache for hot files or columns and monitor refresh and space behavior. |
| One-off TB-scale scan remains slow after pruning |
Insufficient scan capacity. |
Use controlled parallel query or Data Lake Accelerator where available. |
| Join order or method is unstable |
Missing or inaccurate external-table statistics. |
Regather DBMS_STATS and compare estimated with actual rows. |
| Catalog table is stale or missing |
DBMS_CATALOG metadata cache is stale. |
Flush or prefill catalog metadata. Do not refresh Lake Cache for a metadata-only problem. |
| Table is visible but cached results lag source files |
Lake Cache lifecycle has not incorporated source changes. |
Refresh or retire cached files according to the policy. |
| High-concurrency workload misses latency target |
Remote I/O and external variability remain on the critical path. |
Materialize or load the required data into Oracle AI Database and add database-native optimization structures. |