postgresql/doc/FAQ_DEV
2002-02-25 20:29:39 +00:00

667 lines
31 KiB
Plaintext

Developer's Frequently Asked Questions (FAQ) for PostgreSQL
Last updated: Mon Feb 25 15:29:28 EST 2002
Current maintainer: Bruce Momjian (pgman@candle.pha.pa.us)
The most recent version of this document can be viewed at the
postgreSQL Web site, http://www.PostgreSQL.org.
_________________________________________________________________
General Questions
1.1) How do I get involved in PostgreSQL development?
1.2) How do I add a feature or fix a bug?
1.3) How do I download/update the current source tree?
1.4) How do I test my changes?
1.5) What tools are available for developers?
1.6) What books are good for developers?
1.7) What is configure all about?
1.8) How do I add a new port?
1.9) Why don't we use threads in the backend?
1.10) How are RPM's packaged?
1.11) How are CVS branches handled?
Technical Questions
2.1) How do I efficiently access information in tables from the
backend code?
2.2) Why are table, column, type, function, view names sometimes
referenced as Name or NameData, and sometimes as char *?
2.3) Why do we use Node and List to make data structures?
2.4) I just added a field to a structure. What else should I do?
2.5) Why do we use palloc() and pfree() to allocate memory?
2.6) What is elog()?
2.7) What is CommandCounterIncrement()?
_________________________________________________________________
General Questions
1.1) How go I get involved in PostgreSQL development?
This was written by Lamar Owen:
2001-06-22
What open source development process is used by the PostgreSQL team?
Read HACKERS for six months (or a full release cycle, whichever is
longer). Really. HACKERS _is_the process. The process is not well
documented (AFAIK -- it may be somewhere that I am not aware of) --
and it changes continually.
What development environment (OS, system, compilers, etc) is required
to develop code?
Developers Corner on the website has links to this information. The
distribution tarball itself includes all the extra tools and documents
that go beyond a good Unix-like development environment. In general, a
modern unix with a modern gcc, GNU make or equivalent, autoconf (of a
particular version), and good working knowledge of those tools are
required.
What areas need support?
The TODO list.
You've made the first step, by finding and subscribing to HACKERS.
Once you find an area to look at in the TODO, and have read the
documentation on the internals, etc, then you check out a current
CVS,write what you are going to write (keeping your CVS checkout up to
date in the process), and make up a patch (as a context diff only) and
send to the PATCHES list, prefereably.
Discussion on the patch typically happens here. If the patch adds a
major feature, it would be a good idea to talk about it first on the
HACKERS list, in order to increase the chances of it being accepted,
as well as toavoid duplication of effort. Note that experienced
developers with a proven track record usually get the big jobs -- for
more than one reason. Also note that PostgreSQL is highly portable --
nonportable code will likely be dismissed out of hand.
Once your contributions get accepted, things move from there.
Typically, you would be added as a developer on the list on the
website when one of the other developers recommends it. Membership on
the steering committee is by invitation only, by the other steering
committee members, from what I have gathered watching froma distance.
I make these statements from having watched the process for over two
years.
To see a good example of how one goes about this, search the archives
for the name 'Tom Lane' and see what his first post consisted of, and
where he took things. In particular, note that this hasn't been _that_
long ago -- and his bugfixing and general deep knowledge with this
codebase is legendary. Take a few days to read after him. And pay
special attention to both the sheer quantity as well as the
painstaking quality of his work. Both are in high demand.
1.2) How do I add a feature or fix a bug?
The source code is over 250,000 lines. Many problems/features are
isolated to one specific area of the code. Others require knowledge of
much of the source. If you are confused about where to start, ask the
hackers list, and they will be glad to assess the complexity and give
pointers on where to start.
Another thing to keep in mind is that many fixes and features can be
added with surprisingly little code. I often start by adding code,
then looking at other areas in the code where similar things are done,
and by the time I am finished, the patch is quite small and compact.
When adding code, keep in mind that it should use the existing
facilities in the source, for performance reasons and for simplicity.
Often a review of existing code doing similar things is helpful.
1.3) How do I download/update the current source tree?
There are several ways to obtain the source tree. Occasional
developers can just get the most recent source tree snapshot from
ftp.postgresql.org. For regular developers, you can use CVS. CVS
allows you to download the source tree, then occasionally update your
copy of the source tree with any new changes. Using CVS, you don't
have to download the entire source each time, only the changed files.
Anonymous CVS does not allows developers to update the remote source
tree, though privileged developers can do this. There is a CVS FAQ on
our web site that describes how to use remote CVS. You can also use
CVSup, which has similarly functionality, and is available from
ftp.postgresql.org.
To update the source tree, there are two ways. You can generate a
patch against your current source tree, perhaps using the make_diff
tools mentioned above, and send them to the patches list. They will be
reviewed, and applied in a timely manner. If the patch is major, and
we are in beta testing, the developers may wait for the final release
before applying your patches.
For hard-core developers, Marc(scrappy@postgresql.org) will give you a
Unix shell account on postgresql.org, so you can use CVS to update the
main source tree, or you can ftp your files into your account, patch,
and cvs install the changes directly into the source tree.
1.4) How do I test my changes?
First, use psql to make sure it is working as you expect. Then run
src/test/regress and get the output of src/test/regress/checkresults
with and without your changes, to see that your patch does not change
the regression test in unexpected ways. This practice has saved me
many times. The regression tests test the code in ways I would never
do, and has caught many bugs in my patches. By finding the problems
now, you save yourself a lot of debugging later when things are
broken, and you can't figure out when it happened.
1.5) What tools are available for developers?
Aside from the User documentation mentioned in the regular FAQ, there
are several development tools available. First, all the files in the
/tools directory are designed for developers.
RELEASE_CHANGES changes we have to make for each release
SQL_keywords standard SQL'92 keywords
backend description/flowchart of the backend directories
ccsym find standard defines made by your compiler
entab converts tabs to spaces, used by pgindent
find_static finds functions that could be made static
find_typedef finds typedefs in the source code
find_badmacros finds macros that use braces incorrectly
make_ctags make vi 'tags' file in each directory
make_diff make *.orig and diffs of source
make_etags make emacs 'etags' files
make_keywords make comparison of our keywords and SQL'92
make_mkid make mkid ID files
mkldexport create AIX exports file
pgindent indents C source files
pgjindent indents Java source files
pginclude scripts for adding/removing include files
unused_oids in pgsql/src/include/catalog
Let me note some of these. If you point your browser at the
file:/usr/local/src/pgsql/src/tools/backend/index.html directory, you
will see few paragraphs describing the data flow, the backend
components in a flow chart, and a description of the shared memory
area. You can click on any flowchart box to see a description. If you
then click on the directory name, you will be taken to the source
directory, to browse the actual source code behind it. We also have
several README files in some source directories to describe the
function of the module. The browser will display these when you enter
the directory also. The tools/backend directory is also contained on
our web page under the title How PostgreSQL Processes a Query.
Second, you really should have an editor that can handle tags, so you
can tag a function call to see the function definition, and then tag
inside that function to see an even lower-level function, and then
back out twice to return to the original function. Most editors
support this via tags or etags files.
Third, you need to get id-utils from:
ftp://alpha.gnu.org/gnu/id-utils-3.2d.tar.gz
ftp://tug.org/gnu/id-utils-3.2d.tar.gz
ftp://ftp.enst.fr/pub/gnu/gnits/id-utils-3.2d.tar.gz
By running tools/make_mkid, an archive of source symbols can be
created that can be rapidly queried like grep or edited. Others prefer
glimpse.
make_diff has tools to create patch diff files that can be applied to
the distribution. This produces context diffs, which is our preferred
format.
Our standard format is to indent each code level with one tab, where
each tab is four spaces. You will need to set your editor to display
tabs as four spaces:
vi in ~/.exrc:
set tabstop=4
set sw=4
more:
more -x4
less:
less -x4
emacs:
M-x set-variable tab-width
or
; Cmd to set tab stops & indenting for working with PostgreSQL code
(c-add-style "pgsql"
'("bsd"
(indent-tabs-mode . t)
(c-basic-offset . 4)
(tab-width . 4)
(c-offsets-alist .
((case-label . +))))
t) ; t = set this mode on
and add this to your autoload list (modify file path in macro):
(setq auto-mode-alist
(cons '("\\`/usr/local/src/pgsql/.*\\.[chyl]\\'" . pgsql-c-mode)
auto-mode-alist))
or
/*
* Local variables:
* tab-width: 4
* c-indent-level: 4
* c-basic-offset: 4
* End:
*/
pgindent will the format code by specifying flags to your operating
system's utility indent. This article describes the value of a
constent coding style.
pgindent is run on all source files just before each beta test period.
It auto-formats all source files to make them consistent. Comment
blocks that need specific line breaks should be formatted as block
comments, where the comment starts as /*------. These comments will
not be reformatted in any way.
pginclude contains scripts used to add needed #include's to include
files, and removed unneeded #include's.
When adding system types, you will need to assign oids to them. There
is also a script called unused_oids in pgsql/src/include/catalog that
shows the unused oids.
1.6) What books are good for developers?
I have four good books, An Introduction to Database Systems, by C.J.
Date, Addison, Wesley, A Guide to the SQL Standard, by C.J. Date, et.
al, Addison, Wesley, Fundamentals of Database Systems, by Elmasri and
Navathe, and Transaction Processing, by Jim Gray, Morgan, Kaufmann
There is also a database performance site, with a handbook on-line
written by Jim Gray at http://www.benchmarkresources.com.
1.7) What is configure all about?
The files configure and configure.in are part of the GNU autoconf
package. Configure allows us to test for various capabilities of the
OS, and to set variables that can then be tested in C programs and
Makefiles. Autoconf is installed on the PostgreSQL main server. To add
options to configure, edit configure.in, and then run autoconf to
generate configure.
When configure is run by the user, it tests various OS capabilities,
stores those in config.status and config.cache, and modifies a list of
*.in files. For example, if there exists a Makefile.in, configure
generates a Makefile that contains substitutions for all @var@
parameters found by configure.
When you need to edit files, make sure you don't waste time modifying
files generated by configure. Edit the *.in file, and re-run configure
to recreate the needed file. If you run make distclean from the
top-level source directory, all files derived by configure are
removed, so you see only the file contained in the source
distribution.
1.8) How do I add a new port?
There are a variety of places that need to be modified to add a new
port. First, start in the src/template directory. Add an appropriate
entry for your OS. Also, use src/config.guess to add your OS to
src/template/.similar. You shouldn't match the OS version exactly. The
configure test will look for an exact OS version number, and if not
found, find a match without version number. Edit src/configure.in to
add your new OS. (See configure item above.) You will need to run
autoconf, or patch src/configure too.
Then, check src/include/port and add your new OS file, with
appropriate values. Hopefully, there is already locking code in
src/include/storage/s_lock.h for your CPU. There is also a
src/makefiles directory for port-specific Makefile handling. There is
a backend/port directory if you need special files for your OS.
1.9) Why don't we use threads in the backend?
There are several reasons threads are not used:
* Historically, threads were unsupported and buggy.
* An error in one backend can corrupt other backends.
* Speed improvements using threads are small compared to the
remaining backend startup time.
* The backend code would be more complex.
1.10) How are RPM's packaged?
This was written by Lamar Owen:
2001-05-03
As to how the RPMs are built -- to answer that question sanely
requires me to know how much experience you have with the whole RPM
paradigm. 'How is the RPM built?' is a multifaceted question. The
obvious simple answer is that I maintain:
1. A set of patches to make certain portions of the source tree
'behave' in the different environment of the RPMset;
2. The initscript;
3. Any other ancilliary scripts and files;
4. A README.rpm-dist document that tries to adequately document both
the differences between the RPM build and the WHY of the
differences, as well as useful RPM environment operations (like,
using syslog, upgrading, getting postmaster to start at OS boot,
etc);
5. The spec file that throws it all together. This is not a trivial
undertaking in a package of this size.
I then download and build on as many different canonical distributions
as I can -- currently I am able to build on Red Hat 6.2, 7.0, and 7.1
on my personal hardware. Occasionally I receive opportunity from
certain commercial enterprises such as Great Bridge and PostgreSQL,
Inc. to build on other distributions.
I test the build by installing the resulting packages and running the
regression tests. Once the build passes these tests, I upload to the
postgresql.org ftp server and make a release announcement. I am also
responsible for maintaining the RPM download area on the ftp site.
You'll notice I said 'canonical' distributions above. That simply
means that the machine is as stock 'out of the box' as practical --
that is, everything (except select few programs) on these boxen are
installed by RPM; only official Red Hat released RPMs are used (except
in unusual circumstances involving software that will not alter the
build -- for example, installing a newer non-RedHat version of the Dia
diagramming package is OK -- installing Python 2.1 on the box that has
Python 1.5.2 installed is not, as that alters the PostgreSQL build).
The RPM as uploaded is built to as close to out-of-the-box pristine as
is possible. Only the standard released 'official to that release'
compiler is used -- and only the standard official kernel is used as
well.
For a time I built on Mandrake for RedHat consumption -- no more.
Nonstandard RPM building systems are worse than useless. Which is not
to say that Mandrake is useless! By no means is Mandrake useless --
unless you are building Red Hat RPMs -- and Red Hat is useless if
you're trying to build Mandrake or SuSE RPMs, for that matter. But I
would be foolish to use 'Lamar Owen's Super Special RPM Blend Distro
0.1.2' to build for public consumption! :-)
I _do_ attempt to make the _source_ RPM compatible with as many
distributions as possible -- however, since I have limited resources
(as a volunteer RPM maintainer) I am limited as to the amount of
testing said build will get on other distributions, architectures, or
systems.
And, while I understand people's desire to immediately upgrade to the
newest version, realize that I do this as a side interest -- I have a
regular, full-time job as a broadcast
engineer/webmaster/sysadmin/Technical Director which occasionally
prevents me from making timely RPM releases. This happened during the
early part of the 7.1 beta cycle -- but I believe I was pretty much on
the ball for the Release Candidates and the final release.
I am working towards a more open RPM distribution -- I would dearly
love to more fully document the process and put everything into CVS --
once I figure out how I want to represent things such as the spec file
in a CVS form. It makes no sense to maintain a changelog, for
instance, in the spec file in CVS when CVS does a better job of
changelogs -- I will need to write a tool to generate a real spec file
from a CVS spec-source file that would add version numbers, changelog
entries, etc to the result before building the RPM. IOW, I need to
rethink the process -- and then go through the motions of putting my
long RPM history into CVS one version at a time so that version
history information isn't lost.
As to why all these files aren't part of the source tree, well, unless
there was a large cry for it to happen, I don't believe it should.
PostgreSQL is very platform-agnostic -- and I like that. Including the
RPM stuff as part of the Official Tarball (TM) would, IMHO, slant that
agnostic stance in a negative way. But maybe I'm too sensitive to
that. I'm not opposed to doing that if that is the consensus of the
core group -- and that would be a sneaky way to get the stuff into CVS
:-). But if the core group isn't thrilled with the idea (and my
instinct says they're not likely to be), I am opposed to the idea --
not to keep the stuff to myself, but to not hinder the
platform-neutral stance. IMHO, of course.
Of course, there are many projects that DO include all the files
necessary to build RPMs from their Official Tarball (TM).
1.11) How are CVS branches managed?
This was written by Tom Lane:
2001-05-07
If you just do basic "cvs checkout", "cvs update", "cvs commit", then
you'll always be dealing with the HEAD version of the files in CVS.
That's what you want for development, but if you need to patch past
stable releases then you have to be able to access and update the
"branch" portions of our CVS repository. We normally fork off a branch
for a stable release just before starting the development cycle for
the next release.
The first thing you have to know is the branch name for the branch you
are interested in getting at. To do this, look at some long-lived
file, say the top-level HISTORY file, with "cvs status -v" to see what
the branch names are. (Thanks to Ian Lance Taylor for pointing out
that this is the easiest way to do it.) Typical branch names are:
REL7_1_STABLE
REL7_0_PATCHES
REL6_5_PATCHES
OK, so how do you do work on a branch? By far the best way is to
create a separate checkout tree for the branch and do your work in
that. Not only is that the easiest way to deal with CVS, but you
really need to have the whole past tree available anyway to test your
work. (And you *better* test your work. Never forget that dot-releases
tend to go out with very little beta testing --- so whenever you
commit an update to a stable branch, you'd better be doubly sure that
it's correct.)
Normally, to checkout the head branch, you just cd to the place you
want to contain the toplevel "pgsql" directory and say
cvs ... checkout pgsql
To get a past branch, you cd to whereever you want it and say
cvs ... checkout -r BRANCHNAME pgsql
For example, just a couple days ago I did
mkdir ~postgres/REL7_1
cd ~postgres/REL7_1
cvs ... checkout -r REL7_1_STABLE pgsql
and now I have a maintenance copy of 7.1.*.
When you've done a checkout in this way, the branch name is "sticky":
CVS automatically knows that this directory tree is for the branch,
and whenever you do "cvs update" or "cvs commit" in this tree, you'll
fetch or store the latest version in the branch, not the head version.
Easy as can be.
So, if you have a patch that needs to apply to both the head and a
recent stable branch, you have to make the edits and do the commit
twice, once in your development tree and once in your stable branch
tree. This is kind of a pain, which is why we don't normally fork the
tree right away after a major release --- we wait for a dot-release or
two, so that we won't have to double-patch the first wave of fixes.
Technical Questions
2.1) How do I efficiently access information in tables from the backend code?
You first need to find the tuples(rows) you are interested in. There
are two ways. First, SearchSysCache() and related functions allow you
to query the system catalogs. This is the preferred way to access
system tables, because the first call to the cache loads the needed
rows, and future requests can return the results without accessing the
base table. The caches use system table indexes to look up tuples. A
list of available caches is located in
src/backend/utils/cache/syscache.c.
src/backend/utils/cache/lsyscache.c contains many column-specific
cache lookup functions.
The rows returned are cache-owned versions of the heap rows.
Therefore, you must not modify or delete the tuple returned by
SearchSysCache(). What you should do is release it with
ReleaseSysCache() when you are done using it; this informs the cache
that it can discard that tuple if necessary. If you neglect to call
ReleaseSysCache(), then the cache entry will remain locked in the
cache until end of transaction, which is tolerable but not very
desirable.
If you can't use the system cache, you will need to retrieve the data
directly from the heap table, using the buffer cache that is shared by
all backends. The backend automatically takes care of loading the rows
into the buffer cache.
Open the table with heap_open(). You can then start a table scan with
heap_beginscan(), then use heap_getnext() and continue as long as
HeapTupleIsValid() returns true. Then do a heap_endscan(). Keys can be
assigned to the scan. No indexes are used, so all rows are going to be
compared to the keys, and only the valid rows returned.
You can also use heap_fetch() to fetch rows by block number/offset.
While scans automatically lock/unlock rows from the buffer cache, with
heap_fetch(), you must pass a Buffer pointer, and ReleaseBuffer() it
when completed.
Once you have the row, you can get data that is common to all tuples,
like t_self and t_oid, by merely accessing the HeapTuple structure
entries. If you need a table-specific column, you should take the
HeapTuple pointer, and use the GETSTRUCT() macro to access the
table-specific start of the tuple. You then cast the pointer as a
Form_pg_proc pointer if you are accessing the pg_proc table, or
Form_pg_type if you are accessing pg_type. You can then access the
columns by using a structure pointer:
((Form_pg_class) GETSTRUCT(tuple))->relnatts
You must not directly change live tuples in this way. The best way is
to use heap_modifytuple() and pass it your original tuple, and the
values you want changed. It returns a palloc'ed tuple, which you pass
to heap_replace(). You can delete tuples by passing the tuple's t_self
to heap_destroy(). You use t_self for heap_update() too. Remember,
tuples can be either system cache copies, which may go away after you
call ReleaseSysCache(), or read directly from disk buffers, which go
away when you heap_getnext(), heap_endscan, or ReleaseBuffer(), in the
heap_fetch() case. Or it may be a palloc'ed tuple, that you must
pfree() when finished.
2.2) Why are table, column, type, function, view names sometimes referenced
as Name or NameData, and sometimes as char *?
Table, column, type, function, and view names are stored in system
tables in columns of type Name. Name is a fixed-length,
null-terminated type of NAMEDATALEN bytes. (The default value for
NAMEDATALEN is 32 bytes.)
typedef struct nameData
{
char data[NAMEDATALEN];
} NameData;
typedef NameData *Name;
Table, column, type, function, and view names that come into the
backend via user queries are stored as variable-length,
null-terminated character strings.
Many functions are called with both types of names, ie. heap_open().
Because the Name type is null-terminated, it is safe to pass it to a
function expecting a char *. Because there are many cases where
on-disk names(Name) are compared to user-supplied names(char *), there
are many cases where Name and char * are used interchangeably.
2.3) Why do we use Node and List to make data structures?
We do this because this allows a consistent way to pass data inside
the backend in a flexible way. Every node has a NodeTag which
specifies what type of data is inside the Node. Lists are groups of
Nodes chained together as a forward-linked list.
Here are some of the List manipulation commands:
lfirst(i)
return the data at list element i.
lnext(i)
return the next list element after i.
foreach(i, list)
loop through list, assigning each list element to i. It is
important to note that i is a List *, not the data in the List
element. You need to use lfirst(i) to get at the data. Here is
a typical code snippet that loops through a List containing Var
*'s and processes each one:
List *i, *list;
foreach(i, list)
{
Var *var = lfirst(i);
/* process var here */
}
lcons(node, list)
add node to the front of list, or create a new list with node
if list is NIL.
lappend(list, node)
add node to the end of list. This is more expensive that lcons.
nconc(list1, list2)
Concat list2 on to the end of list1.
length(list)
return the length of the list.
nth(i, list)
return the i'th element in list.
lconsi, ...
There are integer versions of these: lconsi, lappendi, nthi.
List's containing integers instead of Node pointers are used to
hold list of relation object id's and other integer quantities.
You can print nodes easily inside gdb. First, to disable output
truncation when you use the gdb print command:
(gdb) set print elements 0
Instead of printing values in gdb format, you can use the next two
commands to print out List, Node, and structure contents in a verbose
format that is easier to understand. List's are unrolled into nodes,
and nodes are printed in detail. The first prints in a short format,
and the second in a long format:
(gdb) call print(any_pointer)
(gdb) call pprint(any_pointer)
The output appears in the postmaster log file, or on your screen if
you are running a backend directly without a postmaster.
2.4) I just added a field to a structure. What else should I do?
The structures passing around from the parser, rewrite, optimizer, and
executor require quite a bit of support. Most structures have support
routines in src/backend/nodes used to create, copy, read, and output
those structures. Make sure you add support for your new field to
these files. Find any other places the structure may need code for
your new field. mkid is helpful with this (see above).
2.5) Why do we use palloc() and pfree() to allocate memory?
palloc() and pfree() are used in place of malloc() and free() because
we find it easier to automatically free all memory allocated when a
query completes. This assures us that all memory that was allocated
gets freed even if we have lost track of where we allocated it. There
are special non-query contexts that memory can be allocated in. These
affect when the allocated memory is freed by the backend.
2.6) What is elog()?
elog() is used to send messages to the front-end, and optionally
terminate the current query being processed. The first parameter is an
elog level of DEBUG (levels 1-5), LOG, INFO, NOTICE, ERROR, FATAL, or
PANIC. NOTICE prints on the user's terminal and the postmaster logs.
INFO prints only to the user's terminal and LOG prints only to the
server logs. (These can be changed from postgresql.conf.) ERROR prints
in both places, and terminates the current query, never returning from
the call. FATAL terminates the backend process. The remaining
parameters of elog are a printf-style set of parameters to print.
elog(ERROR) frees most memory and open file descriptors so you don't
need to clean these up before the call.
2.7) What is CommandCounterIncrement()?
Normally, transactions can not see the rows they modify. This allows
UPDATE foo SET x = x + 1 to work correctly.
However, there are cases where a transactions needs to see rows
affected in previous parts of the transaction. This is accomplished
using a Command Counter. Incrementing the counter allows transactions
to be broken into pieces so each piece can see rows modified by
previous pieces. CommandCounterIncrement() increments the Command
Counter, creating a new part of the transaction.