Обсуждение: Clarification on two bits on VACUUM FULL VERBOSE output
Folks, Wanted to get clarification on two bits of output from 7.4's VACUUM FULL VERBOSE: "Total free space (including removable row versions) is 2932036 bytes." If the table referenced has no dead row versions, does this indicate open space on partially full pages? "There were 33076 unused item pointers." Is this a count of dead index pointers, or something else? -- --Josh Josh Berkus Aglio Database Solutions San Francisco
Josh Berkus <josh@agliodbs.com> writes: > Wanted to get clarification on two bits of output from 7.4's VACUUM FULL > VERBOSE: > "Total free space (including removable row versions) is 2932036 bytes." > If the table referenced has no dead row versions, does this indicate open > space on partially full pages? Yes. > "There were 33076 unused item pointers." > Is this a count of dead index pointers, or something else? No, it's currently-unused item pointers (a/k/a line pointers) on heap pages. See http://developer.postgresql.org/docs/postgres/page.html regards, tom lane
Tom, > > "There were 33076 unused item pointers." > > Is this a count of dead index pointers, or something else? > > No, it's currently-unused item pointers (a/k/a line pointers) on heap > pages. See http://developer.postgresql.org/docs/postgres/page.html So this would be a count of pointers whose items had already been moved? -- --Josh Josh Berkus Aglio Database Solutions San Francisco
Josh Berkus <josh@agliodbs.com> writes: >>> "There were 33076 unused item pointers." >>> Is this a count of dead index pointers, or something else? >> >> No, it's currently-unused item pointers (a/k/a line pointers) on heap >> pages. See http://developer.postgresql.org/docs/postgres/page.html > So this would be a count of pointers whose items had already been moved? Either deleted, or moved to another page during VACUUM FULL compaction. Such a pointer can be recycled to point to a new item, if there's room to put another item on its page ... but if not, the pointer is wasted space. I don't believe we ever try to physically eliminate unused item pointers. regards, tom lane