Skip to content
Snippets Groups Projects
  1. Apr 16, 2020
  2. Apr 14, 2020
  3. Apr 07, 2020
    • Christoph Wurst's avatar
      Lower the memory footprint of the initial message cache sync · 72a5703e
      Christoph Wurst authored
      
      The initial message sync has to fetch potentially large amounts of data
      and insert that into the database. To work around limitations with sync
      requests triggered by web requests the process had already been made
      interruptable and resumable. This means we never insert all the data
      right away. Yet, the IMAP code fetched all UIDs before we capped it to a
      maximum number of results per sync attempt. Depending on the mailbox
      size this operation could require and allocate a lot of memory. On some
      setup with lower memory limits, the process was aborted by the web
      server due to a php memory exhaustion.
      
      This patch modifies the IMAP code to optimize the memory usage by
      limiting the amount of data that is fetched with each initial sync
      attempt. The algorithm works as follows.
      
      IMAP allows us to search in a range with a lower an upper bound UID.
      While we know the highest known UID from the current cache values, we
      can't derive the range for the next page from that as UIDs are not
      continuous but might have holes due to deleted messages. If we assume
      that messages of a mailbox are roughly distributed equally across the
      assigned UIDs we can guess the max UID for the next range.
      So we ask the server for min and max UIDs. The min or our known highest
      UID is always the lower bound. Then we can calculate the distribution
      rate from the min, max and number of messages and build the upper bound.
      On everage this will fetch about the expected number of messages. It
      could be more, but it could also be less. It shouldn't matter in most
      cases.
      
      Signed-off-by: default avatarChristoph Wurst <christoph@winzerhof-wurst.at>
      72a5703e
  4. Mar 20, 2020
  5. Mar 03, 2020
  6. Feb 26, 2020
  7. Feb 11, 2020
  8. Feb 10, 2020
  9. Feb 03, 2020
  10. Jan 31, 2020
Loading