Reply-To: hackers@hub.org, Dan McGuirk <mcguirk@indirect.com>
To: hackers@hub.org
Subject: [HACKERS] tmin writeback optimization
I was doing some profiling of the backend, and noticed that during a certain
benchmark I was running somewhere between 30% and 75% of the backend's CPU
time was being spent in calls to TransactionIdDidCommit() from
HeapTupleSatisfiesNow() or HeapTupleSatisfiesItself() to determine that
changed rows' transactions had in fact been committed even though the rows'
tmin values had not yet been set.
When a query looks at a given row, it needs to figure out whether the
transaction that changed the row has been committed and hence it should pay
attention to the row, or whether on the other hand the transaction is still
in progress or has been aborted and hence the row should be ignored. If
a tmin value is set, it is known definitively that the row's transaction
has been committed. However, if tmin is not set, the transaction
referred to in xmin must be looked up in pg_log, and this is what the
backend was spending a lot of time doing during my benchmark.
So, implementing a method suggested by Vadim, I created the following
patch that, the first time a query finds a committed row whose tmin value
is not set, sets it, and marks the buffer where the row is stored as
dirty. (It works for tmax, too.) This doesn't result in the boost in
real time performance I was hoping for, however it does decrease backend
CPU usage by up to two-thirds in certain situations, so it could be
rather beneficial in high-concurrency settings.
Subject: [HACKERS] linux/alpha patches
These patches lay the groundwork for a Linux/Alpha port. The port doesn't
actually work unless you tweak the linker to put all the pointers in the
first 32 bits of the address space, but it's at least a start. It
implements the test-and-set instruction in Alpha assembly, and also fixes
a lot of pointer-to-integer conversions, which is probably good anyway.
I found another bug in btree index. Looking at the code it seems that NULL
keys are never used to build or scan a btree index (see the explain commands
in the example). However this is not the case when a null key is retrieved
in an outer loop of a join select and used in an index scan of an inner loop.
This bug causes at least three kinds of problems:
1) the backend crashes when it tries to compare a text string with a null.
2) it is not possible to find tuples with null keys in a join.
3) null is considered equal to 0 when the datum is passed by value, see
the last query.
Submitted by: Massimo Dal Zotto <dz@cs.unitn.it>
- Added the header access/heapam.h.
- Changed all instances of "length" to "data_length" to quiet
the compiler.
- initialized a few variables. The compiler couldn't see that
the code guaranteed that these would be initialized before
being dereferenced. If anyone wants to check my work follow
the usage of these variables and make sure that this true
and wasn't actually a bug in the original code.
- added a missing break statement to a default case. This
was a benign error but bad style.
- layed out heap_sysattrlen differently. I think this way
makes the structure of the code crystal clear. There should
be no actual difference in the actual behaviour of the code.
Submitted by: darcy@druid.druid.com (D'Arcy J.M. Cain)
Here's a small patch that my run-time checker whines about
incessantly. The justification for the patch is along the
lines of passing a NULL is allowed if you have an
arguement that is a *POINTER* to something, but if
the arguement is an array reference, it's not really
a "pointer", so it can't be NULL.
If you question this, I refer you to
<URL:http://www.va.pubnix.com/staff/djm/lore/arrays-are-not-pointers>
Anyways, here's the patch:
-Kurt
Submitted by: "Kurt J. Lidl" <lidl@va.pubnix.com>
Vadim B. Mikheev
Bruce Momjian
Marc G. Fournier
Bryan Henderson