Fix redis lua script errors #259
Conversation
|
Thanks! I'll dig into this on Saturday. |
|
@Bogdanp I may have misread the test failure output, but it looks like the redis tests are failing because it can't persist to disk. Maybe the disk is full? |
| @@ -75,6 +75,42 @@ for i=8,#ARGV do | |||
| ARGS[i - 7] = ARGV[i] | |||
| end | |||
|
|
|||
|
|
|||
| -- In my testing I found the max size that we can unpack to be 7999 (which | |||
| -- is LUAI_MAXCSTACK - 1). Lua can be built with a different LUAI_MAXCSTACK | |||
There was a problem hiding this comment.
Is there any way we can get this exact value at runtime?
There was a problem hiding this comment.
Unfortunately it's only exposed in lua's C api. AFAICT, there isn't a way to get those from lua code.
Most likely, the process is not allowed to write to disk and the test changes that you've made push it past the number of commands (10k by default?) it can run before persisting. I'll see what I can do about it next weekend. Have you started running these changes in production? If so, do they seem to have fixed the problems you ran into? This looks good, but I didn't get a chance to dive in as deep as I would've liked this weekend so I'm going to hold off merging until the next weekend. |
|
I haven't run this in production yet. Instead we temporarily set the maintenance chance to 0 since it was consistently failing for us. We may try running this in production next week, as well as the other PR I put up (which I think has a better chance of fixing our "lazy worker" issue). I did run the new tests I wrote on the old code and verified that it failed, so the fact that they pass for me convinces me it will get the maintenance code working again. ETA: the other PR I mentioned is #262 |
|
Thanks! I made a couple changes on top of this and fixed the CI and the changes are now in master. They'll be release with 1.8. |
Calling
unpackon a list with7999items or more causes lua to throw an error (reason why is in the code comments). This will fix errors of that kind in the lua script (primarily encountered in thedo_maintenancebranch).At least partially fixes #258 (maybe fixes the whole thing; time will tell).