Subject: Re: [HACKERS] abstime "now" broken
Yes, I broke 'now' :( with an attempt at a bug fix involving
servers running in the UTC/GMT timezone. These patches fix
the problem, and have been tested in GMT (+00 hours),
PST (-08), and NZT (+12) timezones which exercized the code for
various cases including across day boundaries. btw, this code
fixes the same type of problem for 'today', 'yesterday', 'tomorrow',
for DATETIME, ABSTIME, DATE and TIME types.
The bugfix itself is quite small, but I have accumulated other
changes in the datetime data type and include them here also.
One set of changes involves printing ISO-formatted dates and
is in response to the helpful information from Kurt Lidl regarding
ANSI SQL dates. I'll send another e-mail sometime soon discussing
more issues he has raised...
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] backend/utils/adt/timestamp.c
Back to this timezone stuff. The struct tm has a field (tm_gmtoff) which
is the offset from UTC (GMT is archaic BTW) in seconds. Is this the
value you are looking for when you use timezone? Note that this applies
to NetBSD but it does not appear to be in either ANSI C or POSIX. This
looks like one of those things that is just going to have to be hand
coded for each platform.
Why not just store the values in UTC and use localtime instead of
gmtime when retrieving the value?
Also, you assume the time is returned as a 4 byte integer. In fact,
there is not even any requirement that time be an integral value. You
should use time_t here.
The input function seems unduly restrictive. Somewhere in the sources
there is an input function that allows words for months. Can't we do
the same here?
There is a standard function, difftime, for subtracting two times. It
deals with cases where time_t is not integral. There is, however, a
small performance hit since it returns a double and I don't believe
there is any system currently which uses anything but an integral for
time_t. Still, this is technically the correct and portable thing to do.
The returns from the various comparisons should probably be a bool.
of endian.h. I figure that if it exists it's pretty sure that it has
the byte order information and we may catch some other ports without
any further testing.
From: "D'Arcy J.M. Cain" <darcy@druid.net>
Subject: [HACKERS] More patches for date/time
I have accumulated several patches to add functionality to the datetime
and timespan data types as well as to fix reported porting bugs on non-BSD
machines. These patches are:
dt.c.patch - add datetime_part(), fix bugs
dt.h.patch - add quarter and timezone support, add prototypes
globals.c.patch - add time and timezone variables
miscadmin.h.patch - add time and timezone variables
nabstime.c.patch - add datetime conversion routine
nabstime.h.patch - add prototypes
pg_operator.h.patch - add datetime operators, clean up formatting
pg_proc.h.patch - add datetime functions, reassign conflicting date OIDs
pg_type.h.patch - add datetime and timespan data types
The dt.c and pg_proc.h patches are fairly large; the latter mostly because I tried
to get some columns for existing entries to line up.
Subject: [HACKERS] libpq/pqcomm stuff and Solaris byte order
I decided to go ahead with the required changes since no one else seems
to. I don't guarantee that it is perfect but with these changes the
package actually compiles. While I was at it I added to the Sparc
Solaris header to define the byte order. Note that NetBSD sets this
in the system headers so it wasn't required there.
In particular, someone may want to check whether I removed the correct
84 lines from backend/libpq/pqcomprim.c.
of common routines in pqcomprim.c (pq communication primitives).
Not all adapted to it yet, but it's a start.
- Rewritten some of those routines, to write/read bigger chunks of
data, precomputing stuff in buffers instead of sending out byte
by byte.
- As a consequence, I need to know the endianness of the machine.
Currently I rely on getting it from machine/endian.h, but this
may not be available everywhere? (Who the hell thought it was
a good idea to pass integers to the backend the other way around
than the normal network byte order? *argl*)
- Libpq looks in the environment for magic variables, and upon
establishing a connection to the backend, sends it queries
of the form "SET var_name TO 'var_value'". This needs a change
in the backend parser (Mr. Parser, are you there? :)
- Currently it looks for two Env-Vars, namely PG_DATEFORMAT
and PG_FLOATFORMAT. What else makes sense? PG_TIMEFORMAT?
PG_TIMEZONE?
From: "Martin J. Laubach" <mjl@wwx.vip.at>
Subject: [HACKERS] Patch for io routines
I am currently trying to improve on the front-backend communication
routines; and noticed that lots of code are duplicated for libpq and
the backend. This is a first patch that tries to share code between
the two, more to follow.
mjl
Subject: [HACKERS] password authentication
This patch adds support for plaintext password authentication. To use
it, you add a line like
host all 0.0.0.0 0.0.0.0 password pg_pwd.conf
to your pg_hba.conf, where 'pg_pwd.conf' is the name of a file containing
the usernames and password hashes in the format of the first two fields
of a Unix /etc/passwd file. (Of course, you can use a specific database
name or IP instead.)
Then, to connect with a password through libpq, you use the PQconnectdb()
function, specifying the "password=" tag in the connect string and also
adding the tag "authtype=password".
I also added a command-line switch '-u' to psql that tells it to prompt
for a username and password and use password authentication.
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.
Subject: [HACKERS] better access control error messages
This patch replaces the 'no such class or insufficient privilege' with
distinct error messages that tell you whether the table really doesn't
exist or whether access was denied.
The following patches add to the backend a new debugging flag -K which prints
a debug trace of all locking operations on user relations (those with oid
greater than 20000). The code is compiled only if LOCK_MGR_DEBUG is defined,
so the patch should be harmless if not explicitly enabled.
I'm using the code to trace deadlock conditions caused by application queries
using the command "$POSTMASTER -D $PGDATA -o '-d 1 -K 1'.
The patches are for version 6.0 dated 970126.
Patches from: aoki@CS.Berkeley.EDU (Paul M. Aoki)
i gave jolly my btree bulkload code a long, long time ago but never
gave him a bunch of my bugfixes. here's a diff against the 6.0
baseline.
for some reason, this code has slowed down somewhat relative to the
insertion-build code on very small tables. don't know why -- it used
to be within about 10%. anyway, here are some (highly unscientific!)
timings on a dec 3000/300 for synthetic tables with 10k, 100k and
1000k tuples (basically, 1mb, 10mb and 100mb heaps). 'c' means
clustered (pre-sorted) inputs and 'u' means unclustered (randomly
ordered) inputs. the 10k table basically fits in the buffer pool, but
the 100k and 1000k tables don't. as you can see, insertion build is
fine if you've sorted your heaps on your index key or if your heap
fits in core, but is absolutely horrible on unordered data (yes,
that's 7.5 hours to index 100mb of data...) because of the zillions of
random i/os.
if it doesn't work for you for whatever reason, you can always turn it
back off by flipping the FastBuild flag in nbtree.c. i don't have
time to maintain it.
good luck!
baseline code:
time psql -c 'create index c10 on k10 using btree (c int4_ops)' bttest
real 8.6
time psql -c 'create index u10 on k10 using btree (b int4_ops)' bttest
real 9.1
time psql -c 'create index c100 on k100 using btree (c int4_ops)' bttest
real 59.2
time psql -c 'create index u100 on k100 using btree (b int4_ops)' bttest
real 652.4
time psql -c 'create index c1000 on k1000 using btree (c int4_ops)' bttest
real 636.1
time psql -c 'create index u1000 on k1000 using btree (b int4_ops)' bttest
real 26772.9
bulkloading code:
time psql -c 'create index c10 on k10 using btree (c int4_ops)' bttest
real 11.3
time psql -c 'create index u10 on k10 using btree (b int4_ops)' bttest
real 10.4
time psql -c 'create index c100 on k100 using btree (c int4_ops)' bttest
real 59.5
time psql -c 'create index u100 on k100 using btree (b int4_ops)' bttest
real 63.5
time psql -c 'create index c1000 on k1000 using btree (c int4_ops)' bttest
real 636.9
time psql -c 'create index u1000 on k1000 using btree (b int4_ops)' bttest
real 701.0
Essentially, config.h now includes an 'os.h', which is created via
configure by linking a "port.h" file from the port directory to the
include directory.
Going to try to merge backend/port in similar ways