| Previous | Next |
| DB_E_NOCOMMAND | DB_E_BADBOOKMARK |
DB_E_COSTLIMIT
Exact value and interpretation
DB_E_COSTLIMIT has the unsigned 32-bit value 2147749389 (0x80040E0D) and the signed representation -2147217907. AllStat defines the result as “Query plan within the cost limit cannot be found”. In the concrete failure represented here, the provider cannot find an execution plan that satisfies the cost limits accepted for the command.
The high bit is set for it, so it is a failure HRESULT rather than a success or informational status. Its facility field is 4 (FACILITY_ITF) and its low code is 3597 (0x0E0D). Those bit fields place this result in an interface-defined family, but they do not reveal the provider, object identity, method, rowset generation or command state that produced it.
OLE DB contract boundary
This result must be interpreted against this contract: an OLE DB command moves through explicit states: command definition, properties and parameters are established before optional preparation and execution; dialect, query-tree and optimizer contracts are provider-specific and should be inferred only from reported capabilities.
Start with the active command object, its state transition history, dialect and provider error records when investigating this result. Preserve it before ADO, ATL,.NET, a database abstraction layer or an application exception replaces it with a generic message; the exact interface and method matter because one OLE DB object can expose several contracts with different preconditions.
Diagnostic sequence
- Capture this result immediately at the native OLE DB return and obtain the current OLE DB error object before another COM call replaces thread error information.
- Identify the exact stage for it: the provider cannot find an execution plan that satisfies the cost limits accepted for the command.
- compare the live command, rowset, accessor or schema state with the metadata and properties actually granted by the provider.
- inspect per-binding, per-property, per-row or per-record statuses whenever the method supplies them; the aggregate result may not identify the rejected element.
- reproduce the issue with the smallest command, rowset or definition operation that preserves the same contract boundary.
- apply one evidence-backed correction, then verify that the operation succeeds and does not merely change into a nearby HRESULT.
Specific conditions that produce it
- Cause 1 for it: the memory, time or resource ceiling is lower than every viable plan.
- Cause 2 for it: statistics or indexes make all candidate plans too expensive.
- Cause 3 for it: combined cost goals constrain the optimizer beyond provider capability.
Evidence to collect before changing the system
A useful it record includes provider CLSID and version, process architecture, interface and method, COM apartment and thread, object correlation ID, transaction state, and the first preceding HRESULT. When it involves command text, parameter values, object names and query-plan details, record types, lengths, hashes or redacted identifiers instead of secrets or full business data.
- Evidence 1 for it: the accepted cost-limit values and optimization goals.
- Evidence 2 for it: the command shape, parameter metadata and provider statistics version.
- Evidence 3 for it: provider error records or plan diagnostics explaining the limiting resource.
Corrective actions
- Action 1 for it: relax the specific limit supported by evidence.
- Action 2 for it: improve indexes, statistics or command predicates to reduce plan cost.
- Action 3 for it: avoid silently removing limits that enforce an operational safety boundary.
Retry and recovery policy
Retry rule for it: retry only after the plan inputs or accepted limits change. A safe it retry must use a changed input, object generation, provider capability or state transition. If the call returning it could have created, updated, deleted or copied data, determine partial completion before replaying it.
Use bounded retries and preserve cancellation. Configuration and contract failures should normally fail fast; concurrency, resource or transient state failures may justify retry only after their stated precondition changes.
Difference from related HRESULT values
DB_E_ABORTLIMITREACHED reports execution stopped after a resource limit was reached, while it reports failure to find a plan within the limit before results are produced. Keep these outcomes separate in telemetry and user-facing remediation because it requires a different next action.
Practical incident
A reporting query receives a strict row-processing limit after a table grows and no plan fits; adding a selective index allows execution without discarding the guardrail and clears it. The diagnostic value comes from retaining it together with the failing interface and object state, not from reducing every provider result to “database error”.
Implementation guidance
Code handling it should release OLE DB resources in ownership order, preserve every provider error record, and log granted properties rather than only requested properties. When handling it, handles such as HACCESSOR, HROW, HCHAPTER and provider-specific region tokens must never be treated as portable integers across object lifetimes.
When it crosses an abstraction boundary, attach a stable correlation ID and structured fields for the native HRESULT, provider source, interface IID, method, object generation and operation phase. Do not log passwords, access tokens, complete SQL text or unrestricted row values merely to make the event easier to search.
Official Microsoft references
Looking for a different code? Search another status or error code.