Last modified: 2010-05-15 15:38:17 UTC
Overview Description: After a clicked rollback (on a diff page) seems to time out (1-2+ minutes without a response), and despite checking the history (in a different window) to see that the rollback did not register, reloading the diff page (in the same new window) and clicking rollback again produces multiple reverts in the article history, which seem to duplicate themselves. Steps to Reproduce (attempts to reproduce failed, so here is a structured rundown of what led to it): 1) Load diff from RC in a new window, click rollback. 2) After waiting for the rollback to register for 1-2 minutes, open target article's history in a new window, to see if the revert has gone through. This obviously requires lag, which is unpredictable. 3) After confirming that the rollback did not go through, load the current diff from the history page and click rollback again. 4) Rinse and repeat, using the "rollback" links already available in the two windows. I actually only did the re-clicking twice, for a total of three rollbacks. Actual Results: Once the server finally responds to the rollback requests, each window will load a page with the title "Action complete" and the body text "Reverted edits by X to last version by X" in large point. Revisiting the history shows that all three reverts went through, separated by as much as two minutes. Additionally, each revert seems to be either duplicated or quadruplicated, resulting in ten consecutive reverts of the same target edit. These reverts also appear in user contribs. The article text itself was not duplicated. Expected Results: Only the first successful revert should have registered, and all other attempts should have returned a "Rollback failed" page with the explanation that the last editor was not the one you were trying to revert. Attempts to duplicate this bug with quick, simultaneous rollbacks of the same edit from different windows returned the expected failure message, but in these cases no lag was observed. Build Date & Platform: ? - This happened on en.wikipedia.org, 2005-06-30
Haven't seen this happen again, and inspection shows that rollback code (at least now) is reading from the master, so this shouldn't happen.