Обсуждение: Odd behavior in regression test?
Hi. I was able to receive mail yesterday (and mostly got caught up), but am off the air again, probably at least through the weekend. So if this has already been brought up, sorry... I've been working on more patches for automatic type conversion, and they fall on top of some of the recent "junkfilter" patches. So I've been waiting for that to settle down before finalizing my patches, particularly since I've been off the air for e-mail for the last week :( Anyway, the current behavior from the source tree is that the new "junkfilter" regression test fails (as does the "random" test, but I haven't looked into that yet). However, not only does the regression test fail by crashing the backend, it seems to take the postmaster with it. This seems to happen repeatably on my i686/linux system with the current source tree as well as with the same tree with my patches. The fact that the junkfilter test crashes doesn't bother me much (since I'm already working around there I can probably track it down) but the postmaster getting taken out is more worrisome. Is it possible that the recent change from fork/exec to just fork leaves the postmaster more exposed? I can imagine that it might, but don't have any direct experience with it so am just guessing. Any other ideas? Do people see this on other platforms? This is the first time I can recall seeing the postmaster go away on a crash of a backend (but of course my memory isn't what it should be :) - Tom
> > Hi. I was able to receive mail yesterday (and mostly got caught up), but > am off the air again, probably at least through the weekend. So if this > has already been brought up, sorry... > > I've been working on more patches for automatic type conversion, and > they fall on top of some of the recent "junkfilter" patches. So I've > been waiting for that to settle down before finalizing my patches, > particularly since I've been off the air for e-mail for the last week :( > > Anyway, the current behavior from the source tree is that the new > "junkfilter" regression test fails (as does the "random" test, but I > haven't looked into that yet). However, not only does the regression > test fail by crashing the backend, it seems to take the postmaster with > it. This seems to happen repeatably on my i686/linux system with the > current source tree as well as with the same tree with my patches. The > fact that the junkfilter test crashes doesn't bother me much (since I'm > already working around there I can probably track it down) but the > postmaster getting taken out is more worrisome. Random is now seeded automatically, so you will need to seed it with a fixed value before using it. > > Is it possible that the recent change from fork/exec to just fork leaves > the postmaster more exposed? I can imagine that it might, but don't have > any direct experience with it so am just guessing. Any other ideas? Do > people see this on other platforms? This is the first time I can recall > seeing the postmaster go away on a crash of a backend (but of course my > memory isn't what it should be :) My guess is that the postmaster can no longer restart its backends after one of them aborts. Something I need to check into perhaps. -- Bruce Momjian | 830 Blythe Avenue maillist@candle.pha.pa.us | Drexel Hill, Pennsylvania 19026 + If your life is a hard drive, | (610) 353-9879(w) + Christ can be your backup. | (610) 853-3000(h)
> > > > Is it possible that the recent change from fork/exec to just fork leaves > > the postmaster more exposed? I can imagine that it might, but don't have > > any direct experience with it so am just guessing. Any other ideas? Do > > people see this on other platforms? This is the first time I can recall > > seeing the postmaster go away on a crash of a backend (but of course my > > memory isn't what it should be :) > > My guess is that the postmaster can no longer restart its backends after > one of them aborts. Something I need to check into perhaps. > I just tried killing a running backend, and could not get the postmaster to disappear. -- Bruce Momjian | 830 Blythe Avenue maillist@candle.pha.pa.us | Drexel Hill, Pennsylvania 19026 + If your life is a hard drive, | (610) 353-9879(w) + Christ can be your backup. | (610) 853-3000(h)
On Sat, 13 Jun 1998, Bruce Momjian wrote: > > > > > > Is it possible that the recent change from fork/exec to just fork leaves > > > the postmaster more exposed? I can imagine that it might, but don't have > > > any direct experience with it so am just guessing. Any other ideas? Do > > > people see this on other platforms? This is the first time I can recall > > > seeing the postmaster go away on a crash of a backend (but of course my > > > memory isn't what it should be :) > > > > My guess is that the postmaster can no longer restart its backends after > > one of them aborts. Something I need to check into perhaps. > > > > I just tried killing a running backend, and could not get the postmaster > to disappear. Try generating a segmentation fault in a loadable module... works everytime here. -- Peter T Mount peter@retep.org.uk or petermount@earthling.net Main Homepage: http://www.retep.org.uk ************ Someday I may rebuild this signature completely ;-) ************ Work Homepage: http://www.maidstone.gov.uk Work EMail: peter@maidstone.gov.uk
On Sat, 13 Jun 1998, Bruce Momjian wrote: > > Is it possible that the recent change from fork/exec to just fork leaves > > the postmaster more exposed? I can imagine that it might, but don't have > > any direct experience with it so am just guessing. Any other ideas? Do > > people see this on other platforms? This is the first time I can recall > > seeing the postmaster go away on a crash of a backend (but of course my > > memory isn't what it should be :) > > My guess is that the postmaster can no longer restart its backends after > one of them aborts. Something I need to check into perhaps. Yesterday, while I was working on the Large Object Orphaning problem, I was having similar problems. I had to stop and restart the postmaster before I could do anything afterwards. -- Peter T Mount peter@retep.org.uk or petermount@earthling.net Main Homepage: http://www.retep.org.uk ************ Someday I may rebuild this signature completely ;-) ************ Work Homepage: http://www.maidstone.gov.uk Work EMail: peter@maidstone.gov.uk
> > Is it possible that the recent change from fork/exec to just fork leaves > > the postmaster more exposed? I can imagine that it might, but don't have > > any direct experience with it so am just guessing. Any other ideas? Do > > people see this on other platforms? This is the first time I can recall > > seeing the postmaster go away on a crash of a backend (but of course my > > memory isn't what it should be :) > > My guess is that the postmaster can no longer restart its backends after > one of them aborts. Something I need to check into perhaps. Does your postmaster stop running, or does it crash any backend that is started. I am seeing the latter, and the cause appears to be that the postmaster environment after the restart of the shared memory is not proper for a backend. I am looking into it. -- Bruce Momjian | 830 Blythe Avenue maillist@candle.pha.pa.us | Drexel Hill, Pennsylvania 19026 + If your life is a hard drive, | (610) 353-9879(w) + Christ can be your backup. | (610) 853-3000(h)
> > Is it possible that the recent change from fork/exec to just fork leaves > > the postmaster more exposed? I can imagine that it might, but don't have > > any direct experience with it so am just guessing. Any other ideas? Do > > people see this on other platforms? This is the first time I can recall > > seeing the postmaster go away on a crash of a backend (but of course my > > memory isn't what it should be :) > > My guess is that the postmaster can no longer restart its backends after > one of them aborts. Something I need to check into perhaps. > This is now fixed. -- Bruce Momjian | 830 Blythe Avenue maillist@candle.pha.pa.us | Drexel Hill, Pennsylvania 19026 + If your life is a hard drive, | (610) 353-9879(w) + Christ can be your backup. | (610) 853-3000(h)