Changes between Version 7 and Version 8 of Ticket #16296, comment 4


Ignore:
Timestamp:
Jun 17, 2015, 1:40:27 PM (4 years ago)
Author:
leeroy
Comment:

Legend:

Unmodified
Added
Removed
Modified
  • Ticket #16296, comment 4

    v7 v8  
    1919tl;dr How the updater is run cannot be enforced. Therefore, it should be flexible enough for operator choice and it should be fault tolerant. Just in case an operator decides to use some alternative automatic run mechanism (periodic or one-time), or accidentally runs more than one. I, for one operator, do not, under any circumstance, want to re-import data when it takes an hour to import recent (72h), and 12 hours to import a single month of archives.
    2020
    21 If kept, a lock should be created in the updater instance, not the one-time version, or the periodic. Note that mapped memory is used in the patch. A further improvement (see [wiki:bugs.torproject.org/15844 #15844]) would be to remove the use of sockets (I never liked that anyway) and instead use the lock content together with some other validation method. I didn't try anything fancy in the patch because this use of the lock file isn't defined (except to prevent concurrent one-time updater). It's up to you. Please do consider the updated comments in [wiki:bugs.torproject.org/15844 #15844] before deciding to scrap the lock file.
     21If kept, a lock should be created in the updater instance, not the one-time version, or the periodic. Note that mapped memory is used in the patch. A further improvement (see [https://bugs.torproject.org/15844 #15844]) would be to remove the use of sockets (I never liked that anyway) and instead use the lock content together with some other validation method. I didn't try anything fancy in the patch because this use of the lock file isn't defined (except to prevent concurrent one-time updater). It's up to you. Please do consider the updated comments in [https://bugs.torproject.org/15844 #15844] before deciding to scrap the lock file.