| Keybuk | this bug is, err, bugging me | 13:43 |
|---|---|---|
| Keybuk | it looks like four jobs were started from the same event | 13:46 |
| Keybuk | and the one that made it then freed that event, despite the refcount | 13:46 |
| Keybuk | sadmac2: oh, I'm going to beat you | 14:03 |
| sadmac2 | Keybuk: what's going on? Give me the 1 hour of sleep version. | 14:19 |
| Keybuk | you know that "rewind the data and upstart crashes" bug? | 14:24 |
| sadmac2 | yes | 14:27 |
| Keybuk | I think it's caused by a Fedora patch | 14:27 |
| sadmac2 | which? | 14:28 |
| Keybuk | not sure yet | 14:29 |
| Keybuk | but I suspect the re-exec one | 14:29 |
| sadmac2 | well I didn't write that, so its all good | 14:30 |
| sadmac2 | mostly good... | 14:30 |
| Keybuk | just need to prove it | 14:31 |
| Keybuk | basically an inviolate constraint is being violated ;) | 14:32 |
| Keybuk | hmm | 15:55 |
| Keybuk | I can't see anything about your patch that might do that, sadly | 15:55 |
| sadmac2 | have you reproduced this thing? | 15:56 |
| Keybuk | nope | 16:05 |
| Keybuk | though his log just made me go "HUH" | 17:18 |
| Keybuk | I like following state machines on paper | 17:18 |
| Keybuk | it's SO MUCH FUN | 17:18 |
| ion_ | :-) | 17:29 |
| * Keybuk now has ice cream | 17:35 | |
| Keybuk | I have reached the event that kills everything | 17:42 |
| Keybuk | I can see it on the event queue | 17:43 |
| Keybuk | (the bit of paper on the rhs of my desk) | 17:43 |
| Keybuk | Handling started event | 18:06 |
| Keybuk | ? goal changed from stop to start | 18:06 |
| Keybuk | ? state changed from waiting to starting | 18:06 |
| Keybuk | ? respawning too fast, stopped | 18:06 |
| Keybuk | ? respawning too fast, stopped | 18:06 |
| Keybuk | ? goal changed from start to stop | 18:06 |
| Keybuk | ? state changed from starting to waiting | 18:06 |
| Keybuk | ... wonder whether that could be it | 18:06 |
| Keybuk | the double "respawning too fast" is kinda odd | 18:06 |
| Keybuk | but, more than that, the respawn detection is the one thing that could be broken by the date spooling backwards | 18:06 |
| Keybuk | kids | 18:19 |
| Keybuk | remember | 18:19 |
| Keybuk | time(NULL) is not monotonic | 18:19 |
| Keybuk | https://bugs.edge.launchpad.net/upstart/+bug/388873/comments/26 | 18:28 |
| Keybuk | gnnnrgh | 18:37 |
| Keybuk | this is a bug | 18:37 |
| sadmac2 | Keybuk: as opposed to? | 18:38 |
| Keybuk | as in, I can now, following the code, see how it goes wrong :) | 18:39 |
| sadmac2 | oic | 18:49 |
| sadmac2 | can you /fix/ it though? | 18:49 |
| Keybuk | first I write a test case ;) | 18:51 |
| Keybuk | and forward port that to 0.5 to check | 18:51 |
| Keybuk | I'm pretty sure 0.5 is expressly not vulnerable, because I totally rewrote the event queue there | 18:51 |
| Keybuk | interestingly, I appear to have rewritten it to exactly avoid this issue | 18:51 |
| Keybuk | will check the test cases there to see whether I already hit this bug, or a similar one, and didn't realise | 18:51 |
| === Keybuk changed the topic of #upstart to: Upstart 0.5.2 "Something, something, something, D-Bus" | Upstart 0.3.11 "For Friday, June 19th 2009, I'm Jon Masters" | http://upstart.ubuntu.com/ | ||
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!