Re: Backend died while dropping index
От | Steve Wolfe |
---|---|
Тема | Re: Backend died while dropping index |
Дата | |
Msg-id | 002401bfdae0$b64492a0$85755ad1@iboats.com обсуждение исходный текст |
Ответ на | Backend died while dropping index (brianb-pggeneral@edsamail.com) |
Список | pgsql-general |
> 1) Postgres 6.5.3 is a known quantity, and there are more people familiar > with > the common, critical problems, so there are more people who can help solve > problems. > 2) 7.0.2 undoubtedly has some new bugs and problems that are not > apparent yet - bugs that could sink a commercial business that depends > on Postgres (like mine) I'll throw in my two cents, your experience hopefully won't be the same as ours. This week, 6.5.3 somehow horked it's own data quite thoroughly for us. Vacuuming the table would fail miserably - even trying to *describe* the tables shut down the back end. It would complain about a tuple being too large - yet the largest tuple in the table was around 6K. Pg_dump and pg_dumpall would also promptly shut down the back end as soon as they tried to touch the table. Boy, that makes it fun. The cause (we suspect) is from one day when another programmer here did something really dumb, and filled up the disk. The docs for 7.0.2 say that it is supposed to be more reliable in situations like that, and so we decided that it was simply too dangerous to stay with a version that can mangle it's own data beyond repair. So, there are arguments on both sides of the coin. It's critical for your database to be running at all times, so you don't want to bring it down to upgrade. At the same time, the upgrade may buy you a better chance of staying up - or it might break everything. You'll have to make an educated guess, and possibly just flip a coin. And, of course, one of the prime rules of system administration is to make changes reversible. Back up the old version and the data, libraries, interfaces, etc.. If something goes terribly wrong during the upgrade, you can always switch back to the old version and be up and running in just a few minutes.... Ah - Here's one more thing, just in case it benefits anyone. A good upgrade procedure is to take a different machine (even if it's pretty low-end) that is similar in OS and hardware, and install the new version on it first. Copy the data over. *Then* test the things that are likely to break, so that you don't have any surprises. steve
В списке pgsql-general по дате отправления: