| Previous | Next |
| hrwrnDataHasChanged | hrFileOpenReadOnly |
hrKeyChanged
Operational meaning
For hrKeyChanged, this code narrows the investigation to one engine boundary and should not be flattened into a generic exception. The decisive boundary is cursor movement or update caused the current index key to change.
The stored Value is 0x88000652 (positive JET status 1618; 1618). The HRESULT carries a warning, so the call may have useful output or a valid cursor state that must be inspected. Current ESE documentation uses JET_wrnKeyChanged for the corresponding published JET condition.
The first useful distinction is that the record may still exist; the warning says its ordering key changed, not that it was deleted. Start by record old and new key values, index identity, update columns, transaction level, and post-update cursor position. That separates a reproducible incident from a later generic cleanup or service error.
Before altering files or retrying
Preserve the first result before retries, cleanup, service restart, or file replacement changes the evidence. The diagnostic record should identify the exact API phase and the owner of every handle or artifact involved.
- Code-specific observation: record old and new key values, index identity, update columns, transaction level, and post-update cursor position.
- Bookmark and transaction snapshot: record it together with this result and the timestamp of the first occurrence.
- Index and key: record it together with this result and the timestamp of the first occurrence.
- Seek/move flags: record it together with this result and the timestamp of the first occurrence.
- Returned current key: record it together with this result and the timestamp of the first occurrence.
Use hashes, lengths, IDs, generation numbers, and redacted samples instead of copying directory contents or sensitive database values into routine logs. For this it investigation, a complete provenance chain is often more useful than a second automatic retry.
Comparison points
It specifically means that the record may still exist; the warning says its ordering key changed, not that it was deleted. Related values below can appear in the same workflow but require a different response:
hrSeekNotEqual | a less-than-or-equal or greater-than-or-equal seek positioned the cursor without finding an exact key match |
|---|---|
hrTableEmpty | the table opened successfully but contains no current record for navigation |
hrFLDKeyTooBig | an index key exceeded the legacy key budget and was truncated rather than represented in full |
Keep the original constant and hexadecimal value in telemetry. Replacing it with “backup failed” or “database warning” removes the state information needed to select the next legal API call.
Objects and state involved
| Diagnostic layer | cursor position, current-record validity, and ordered index navigation |
|---|---|
| Relevant API surface | seek, move, range, key construction, table open, and post-update navigation |
| Code-specific boundary | cursor movement or update caused the current index key to change |
| Narrow corrective direction | reseek or refresh range state before continuing ordered iteration |
A warning can leave the cursor positioned on a valid non-equal record., cursor state must be rechecked after updates, deletes, and key changes.
Recommended handling
- Record it,
0x88000652, the API name, the current phase, and all live context or file owners. - Verify the decisive condition by record old and new key values, index identity, update columns, transaction level, and post-update cursor position.
- Apply only the narrow correction: reseek or refresh range state before continuing ordered iteration.
- After it, recreate any context invalidated by the failure; do not carry stale HBC, cursor, file, or restore-map state into the retry.
- repeat the smallest non-destructive test that reaches the same boundary, then verify both the return value and the resulting file, cursor, backup, or database state.
Acceptance criteria for a fix
Actions that can make diagnosis worse
- do not assume ordered iteration remains valid after an indexed key changes.
- do not read columns before confirming a current record.
- do not discard the first lower-level Win32, RPC, or JET result merely because a later cleanup call returned a more familiar error.
Technical references
- ESE warning table — API ordering, file semantics, warning/error interpretation, or recovery behavior relevant to it.
- JET_ERR enumeration
- Open-source ESE implementation
Looking for a different code? Search another status or error code.