Re: HELP! My Page Cleaners keep dying!!!

jbo@awips1.abrfc.noaa.gov wrote: > > Running Informix OnLine 5.03 (ancient, I know) and > I continually have page cleaners to go into (E)xit > state and NEVER COME BACK! > > Kinda new to Informix, and I'm not sure how to > figure out why this is happening. The log only > has a "ERROR: page cleaner # X has timed out" > message, and I haven't found a utility which will > give me any more info on it. > > Your suggestions are appreciated!!!! > > Bryan Suggestion # 1: Upgrade to DSA 7.x - page cleaners don't time out there. and they are not processes to fall into the limbo of Unix process states. OnLine threads (as in DSA) do a better limbo. ;-) If you cannot or will not upgrade to DSA, at least get up to 5.6 or higher. However, usually a page-cleaner timeout is a symptom of one or more of the following configuration parameters: 1. Checkpoint interval is too long Example: You have 50,000 buffers and 40,000 are now dirty. You have only 2 chunks. When the checkpoint happens, each page cleaner must be done with its assigned chunk within the 2-minute ceiling. If it does not, it goes into limbo, which only exacerbates (whew!) the problem for other page cleaners. Where that time-limit came from? Why is it not tunable? Why did Coke change their formula? Solution: Shorten the checkpoint interval so that fewer buffers can get dirty. 2. Too many page cleaners If you have 4 huge chunks and lotsa dirty buffers and 20 page cleaners then the page cleaners get in each other's way for the synchronous disk writes. (Can't happen in DSA asynchronous disk IO and/or AIO-VP's). Solution (sometimes) - reduce number of page cleaners. Not a common cause for timeouts. 3. LRU_MAX_DIRTY is too high. As a result, when the checkpoint arrives, there are too *&^%!@ many dirty buffers. Solution: Lower LRU_MAX_DIRTY. I have seen environments with over 100,000 buffers use LRU_MAX_DIRTY of 2, MIN of 0 or 1. That's still a lot of dirty buffers but it's unlikely to result in a 2-minute run for a page-cleaner. Play with the parameters. This means bouncing the system a lot and keeping an eye on the log (use tail -f online.log &). Best of luck. But only if you upgrade to the latest & greatest! Otherwise, no sympathy for lying in your own grave. ;-) -- -- Jake Salomon +-----------------------------------------------------------+ | Diplomacy: The art of getting something off your | | chest without losing your shirt | +------------------------Alfred E. Neuman-------------------+