Expose the Firebird 5 parallel workers and in-place ODS upgrade service options - #23
Open
mariuz wants to merge 4 commits into
Open
Expose the Firebird 5 parallel workers and in-place ODS upgrade service options#23mariuz wants to merge 4 commits into
mariuz wants to merge 4 commits into
Conversation
Firebird 5 can run a backup or restore with several workers, as gbak -par does, but there was no way to ask for it: TIBXBackupService and TIBXRestoreService build their service parameter block without it. Adds a published ParallelWorkers property to TIBXBackupRestoreService, the base both share - isc_spb_bkp_parallel_workers and isc_spb_res_parallel_workers are the same parameter code, so one property serves both directions. The parameter is only sent when it is greater than one and the server is Firebird 5 or later, so nothing changes for existing callers or older servers, which reject the parameter. Zero or one means the default single-threaded behaviour. Requires the isc_spb_bkp_parallel_workers constant, which fbintf does not currently define: MWASoftware/fbintf#8. Verified against a live Firebird 6.0.0 server: with one worker the parameter is not sent; with four it is, the server accepts it, and the resulting backup restores to a working database.
Firebird 5 can upgrade a database's on-disk structure in place, as gfix -upgrade does, instead of requiring a backup and restore. There was no way to ask for it: TIBXValidationService maps its options onto the isc_spb_rpr_* bits and had no entry for it. Adds UpgradeODS to TValidateOption, appended rather than inserted so the existing values keep their ordinals, and sends isc_spb_rpr_upgrade_db only when the server is Firebird 5 or later - an older one rejects the option. Requires the isc_spb_rpr_upgrade_db constant, which fbintf does not currently define. Verified against a live Firebird 6.0.0 server: the assembled repair parameter is 0x1000 and the request completes, leaving a database already at the server's format unchanged.
mariuz
changed the title
isc_dpb_page_size is an integer DPB parameter, but GenerateDPB grouped it with the string ones and wrote it with SetAsString. That stores the digits as text, and the clumplet writer rejects an integer parameter carrying more than four bytes, so DB.Params.Values['page_size'] := '16384'; DB.CreateDatabase; fails with Invalid clumplet buffer structure: length of integer exceeds 4 bytes (5) Only the four-digit page sizes appeared to work, and those worked by accident: TFBAttachment.GenerateCreateDatabaseSQL reads the item back with AsString and pastes it into PAGE_SIZE, so the text happened to round-trip. Setting it as an integer fixes the five-digit sizes and leaves the create path unchanged, since getAsString on an integer item returns IntToStr of its value. Verified against Firebird 6.0.0: creating a database with page_size 16384 now yields MON$DATABASE.MON$PAGE_SIZE = 16384.
TIBTransaction.Create added itself to FTransactionList without checking
it was still there, while Destroy already guarded the matching Remove.
That asymmetry is an access violation on the way out of a GUI program:
EAccessViolation
CREATE, line 1603 of runtime/nongui/IBDatabase.pas
...
FREECOMPONENT / PROCESSASYNCCALLQUEUE
DESTROY, line 137 of include/application.inc
INTERFACES_$$_finalize$, line 38 of interfaces.pas
The LCL frees forms left on Application's async release queue from
inside Application.Destroy, which runs from interfaces.pas finalization
- by which time this unit may already have finalized. Tearing down a
data-aware control there reaches IBX again and constructs a transaction,
and the list it wants to join has gone.
The guard Destroy already had did not work either. Finalization called
Free without nilling, so Assigned stayed true on a dangling pointer and
the test let the code through to read released memory. FreeAndNil is
what makes both guards mean anything.
Left alone deliberately: FCriticalSection.Enter/Leave around
EndTransaction have the same exposure, but pairing a lock with a nil
check is a change to the locking rather than a missing guard, and
nothing has been seen to hit it.
Found from a symbolic backtrace out of Marathon, whose only symptom was
an exit code of 217 after the window had closed and the work was saved.
Verified by reasoning and by a full pass of that project's suites; the
crash itself is intermittent and needs a real window close, so it has
not been watched to disappear.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Firebird 5 can run a backup or restore with several workers, as
gbak -pardoes, but there is currently no way to ask for it:
TIBXBackupServiceandTIBXRestoreServicebuild their service parameter block without it.Change
A published
ParallelWorkersproperty onTIBXBackupRestoreService, the baseboth services share.
isc_spb_bkp_parallel_workersandisc_spb_res_parallel_workersare the same parameter code, so one propertyserves both directions.
The parameter is only added when it is greater than one and the server is
Firebird 5 or later, so nothing changes for existing callers, and an older
server — which rejects the parameter — never sees it. Zero or one means the
present single-threaded behaviour, which is also the default.
Depends on
MWASoftware/fbintf#8, which adds the
isc_spb_bkp_parallel_workersconstant.client/include/consts_pub.incpredates Firebird 5 and does not define it, sothis will not compile without that change.
Verified
Against a live Firebird 6.0.0 server, through
TIBXClientSideBackupService:ParallelWorkers = 1— the parameter is not sent, backup behaves as beforeParallelWorkers = 4—isc_spb_bkp_parallel_workers = 4is sent (checkedwith a temporary trace at the point it is added), the server accepts it, and
the resulting file restores to a working database with all its objects
Found while adding Firebird 4/5/6 support to Marathon, where the option is now
offered on its Database Maintenance dialog.