Re: Add progressive backoff to XactLockTableWait functions

Hi,

After studying proarray and lmgr more closely, I have found several
critical issues in the two patches and underestimated their complexity
and subtlety. Sorry for posting immature patches that may have created
noise.

I now realize that making lock acquisition— (void) LockAcquire(&tag,
ShareLock, false, false); in XactLockTableWait
work on a standby, following Andres’ first suggestion, may be simpler
and require fewer new helpers.

XactLockTableWait fails on a standby simply because we never call
XactLockTableInsert. I am considering invoking XactLockTableInsert
when an XID is added to KnownAssignedXids, and XactLockTableDelete(the
constraint for delete now is not used for main xid in primary) when it
is removed. A preliminary implementation and test shows this approach
kinda works, but still need to confirm it is safe on a standby and
does not break other things.

patched:
Performance counter stats for process id '2331196':

233,947,559 cycles
106,270,044 instructions # 0.45
insn per cycle
12,464,449 cache-misses
9,397 context-switches

60.001191369 seconds time elapsed

head:
Performance counter stats for process id '2435004':

1,258,615,994 cycles
575,248,830 instructions # 0.46
insn per cycle
72,508,027 cache-misses
43,791 context-switches

60.001040589 seconds time elapsed

Best,
Xuneng