Gloda and large IMAP stores

Andrew Sutherland asutherland at
Thu Dec 23 16:21:54 UTC 2010

On 12/23/2010 01:40 AM, Ben Bucksch wrote:
> On 23.12.2010 05:14, Andrew Sutherland wrote:
>> Gloda's correctness relies on hearing all of the modifications that 
>> happen to the message stores and being able to store some of its 
>> states on the message header representations in the .msf files.
> Oh... I see. That sounds problematic, though. What if TB crashes 
> between writing msf and writing gloda?

Gloda defers marking messages indexed in the .msf file "clean"/indexed 
until the global-messages-db.sqlite transaction has hit the disk.  The 
result is that in the event of a crash, the messages will still look 
dirty and the startup folder sweep will take care of them.  The 
PendingCommitTracker is involved with this, if you want to check out the 

>> If Thunderbird roamed the .msf files but not the gloda database
> (FWIW, the proposal was to put all Cache files in Local. That includes 
> IMAPMail. Not sure about Local Folders.)

Yes, that would work much better (apart from local folders).  I think 
that would really only leave the issue of who cleans up that cache if a 
user logs into a machine once, and hopefully the answer is windows?


More information about the tb-planning mailing list