PSMDB-1755 ensure FCBIS terminates correctly in case of errors or shutdown

[igorsol]

Two things are implemented by this PR:

• ensure the backup cursor does not leak in case of errors or shutdown
• ensure that storage engine switched back to original location in case of errors or shutdown. In case when we are going to retry FCBIS attempt this allows new attempt to start with correct storage location

[Copilot]

Pull Request Overview

Ensures FCBIS cleans up correctly on errors/shutdown by closing the backup cursor and restoring the storage location back to the configured dbpath when needed.

• Add _restoreStorageLocation helper and ScopeGuard usage to reliably revert storage location on failure paths.
• Proactively kill any open backup cursor during cancellation to prevent leaks.
• Minor comment fix.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
src/mongo/db/repl/initial_syncer_fcb.h	Declares _restoreStorageLocation to centralize restoring dbpath on failure.
src/mongo/db/repl/initial_syncer_fcb.cpp	Implements storage restoration via ScopeGuard in key callbacks; kills backup cursor during cancellation; fixes a comment typo.

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

Pull Request Overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[ktrushin]

Can we remove the _needToSwitchBackToOriginalDBPath variable (set on line 2291) completely and move this ScopeGuard down on the variable's place?

[igorsol]

You are right, this is possible. I implemented this in an additional commit

[igorsol]

Added two additional commits:

• to remove _needToSwitchBackToOriginalDBPath as @ktrushin suggested
• to avoid deadlock between shutdown thread and storage change activity (more details in the commit's comment)

[Copilot]

Pull Request Overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (2)

src/mongo/db/repl/initial_syncer_fcb.cpp:1

• _switchStorageLocation and _restoreStorageLocation require a GlobalLock (MODE_X), enforced by invariant(opCtx->lockState()->isW()). Here, no Lock::GlobalLock is acquired before calling either function, so this will trip the invariant and/or perform the switch without the required lock. Acquire a GlobalLock before the switch, and also inside the storageGuard before calling _restoreStorageLocation, e.g.: Lock::GlobalLock lk(opCtx.get(), MODE_X); before the unlock/switch, and Lock::GlobalLock lk(opCtx, MODE_X); inside the guard.

/*======

src/mongo/db/repl/initial_syncer_fcb.cpp:2443

• This callback performs a storage location switch under released _mutex but does not set _inStorageChange = true before unlocking. That undermines the shutdown wait mechanism and can reintroduce the deadlock this flag was intended to prevent. Mirror the pattern used in _switchToDownloadedCallback: set _inStorageChange = true before lock.unlock(), and rely on _restoreStorageLocation (invoked by the storageGuard at scope exit) to clear the flag and notify.

        // Switch storage to a dummy location
        lock.unlock();
        status = _switchStorageLocation(opCtx.get(), _cfgDBPath + "/.initialsync/.dummy");
        lock.lock();

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[ktrushin]

LGTM