Re: Hanging processes after database startup

In article <46gh9g$oc5@cssun.mathcs.emory.edu>, Jack Parker <jparker@hpbs3645.boi.hp.com> wrote: >> >Have you guys checked to see if the engine needs a checkpoint? > >We have seen lockups when doing an archive where the engine was waiting for >a tape to be mounted. For a while it runs along merrily, but until the tape >is mounted a checkpoint cannot be performed. If a checkpoint is required >the engine freezes unless it can accomplish one. The nasty thing is that >there are no messages anywhere (that I recall) indicating that this situation >has come to pass. > >Any tbstat command will tell you on the top line whether a CKPT REQD. > >I know its a little thing you've probably already thought of, but you never >know... > >cheers >j. I did a little research over the weekend, and I think that I can better articulate what is happening on my system. Here's one bit of information that I came across that helped me understand what is going on: "System V UNIX systems use semaphone-based queues. If a process cannot acquire a latch, it joins a queue of processes waiting for the latch. This is done by updating a data structure in the database, indicating that it is waiting for a specific latch, and then sleeping on its semaphore. The process relinquishes the CPU, allowing another process to run. When the running process has completed, the waiting process's semaphore is incremented, which causes it to awaken and attempt to retry the latch again. This method minimizes operating system overhead." So what seems to be happening is that these hanging processes seem to be waiting for a latch that never appears. The logical thing to assume is that there is a shared memory problem. The shared memory tunables seem to be set properly, however. Also, we have a OnLine 5.02 UC2 database, much larger than the UC7 databases, that runs fine, without ever having any hanging processes. And finally, the pattern of hanging processes is truly baffling to me. Right after the database is brought up, 50-75% of database processes hang. The number decreases the second day, trickles to just one or two by the third day, and then ceases for the rest of the week, only to restart when the machine is rebooted and the database restarted. Does this seem logical to anyone? Someone in my group pointed out that the root dbspace for the UC7 databases is somewhat small, approximately 10,000 pages, 20 meg. Yet, there appears to be over 6,000 pages of space free, so I don't think that is a factor. And if it were, the problem would get WORSE as more users logged onto the database. The opposite seems to be occuring; the first users are plagued with hangs, and as more people log on, the situation improves. Sorry for the long post. If anyone can shed any light, it would be appreciated. Thanks, Lou Avrami AT&T Law & Govt. Affairs