Updating fails - Database removal generates EBUSY on Windows

[UziTech]

when trying to update php-integrator-base it fails with the error

EBUSY: resource busy or locked, unlink 'C:\Users\tjbrix\.atom\packages\php-integrator-base\indexes\Vic.sqlite-wal

The only way to get it to work is to uninstall php-integrator-base, restart atom then reinstall php-integrator-base

[dipyalov]

I confirm. This happens constantly on windows.

[nahuelhds]

This happens every time I've update it. Then I have to remove ALL the packages and re-install them. Quite annoying haha

Windows 10 64bits

[nahuelhds]

Any news about this? I'd be nice to be able to update without having to uninstall every time. =)

[Gert-dev]

Not yet, but I do have an idea as to why this happens. It is likely the IntiializeCommand in the core that is trying to unlink the database and reinitialize it on startup. It does ensure that the connection is closed before doing that, but the base package kills the old core when shutting down, so the connection likely lingers for a moment and thus generates the busy error. In other words:

• User updates base package.
• Atom deactivates old version of base package down (calls deactivate, which in turn kills the core). Core has no opportunity to close its active connection.
• Atom completes update by replacing base package files.
• Atom activates new version of base package.
• New version of base package downloads the core and starts it.
• Base package tests if project database is outdated, if so, spawns an initialize command.
• Initialize command from core attemps to unlink database to create a new one and fails, because an old connection is still active.

Of course, I haven't actually verified this yet, but this is what I think is happening. A solution would be to not kill the PHP server on deactivation, but rather send a request to ask it to shutdown nicely, then the core could close the connection and exit itself.

[Gert-dev]

The next version is going to include a proper shutdown via a new core exit request, which should fix the issue I mentioned in my last post.

I'm not sure if this will fix all of these kinds of issues, but it should improve the situation in any case.

This will likely be autoclosed after the next version is released, so feel free to create a new ticket if it occurs again afterwards.