- Bacula Regression
- Kern Sibbald
- December 2003
+ Bacula Regression
+ Kern Sibbald
+
+This is Bacula's regression script directory.
+
+!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
+Warning!!!! Make sure not to run it on the same system
+with your production Catalog because the tables will all
+be cleared. You can run it on the your production system
+if you use a different database. E.g. if your production
+system uses MySQL, you can use SQLite here.
+!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
-This is Bacula's regression script directory. At this time
-(December 2003), it is still in development, so all the tests are
-not complete.
To set it up, create your personal configuration file, by
-copying prototype.conf to xxx.conf or simply editing prototype.conf
-directly.
+copying prototype.conf to config or simply editing prototype.conf
+directly then copying it to the file config.
+
+You must end up with a file named config in the main regress
+directory that has all the specifications that correspond to
+your system.
-Then edit your conf file and define appropriate values
-for the variables that are in that file. If you want to see
-a real example, look at kern.conf, but please don't use my
-email address!
+If you are using SQLite, make sure that depkgs is pre-built if it
+isn't already: (cd your-depkgs; make sqlite).
-Make sure that depkgs is pre-built if it isn't
-already: (cd your-depkgs; make sqlite).
+Note, if you use any database other than SQLite, be sure it is not
+your production database because Bacula will delete all the tables
+and recreate them. With SQLite, a new different database is created,
+so it will not affect your production system.
+
+Using the .conf file, you can now select between any Catalog type:
+SQLite, SQLite3, MySQL, or PostgreSQL. Be aware, however, if you
+use an installed database on a production server, running these
+tests will delete all the tables !!!!!!!!!!!!!!!!!! I run my
+tests on a non-production machine, and in addition, I normally use
+SQLite as the database, while my production uses MySQL.
Then do:
- ./config xxx.conf
make setup
You run the above one time. This will build a Makefile from
test. Aside from the required "make setup", each test is totally
self-initalizing and should clean up after itself.
-Not all the tests yet report OK. This is simply because there are
-some spurious differences that I haven't yet taken the time to
-eliminate. The working scrips as of 24 Apr 03 are:
+All the tests expect you to execute them from the main regress
+directory!
-backup-bacula-test
-sparse-test
-compressed-test
-sparse-compressed-test
-two-jobs-test
-wierd-files-test
-verify-vol-test
+You can run all the disk based tests by doing:
-The tests expect you to execute them from the main regress
-directory!
+ ./do_file
+
+The disk based tests are totally separate from any production
+system, provided you have configured the database appropriately
+as noted above.
+
+You can run all the disk and most of the tape tests by doing:
+
+ ./do_all
+
+======== Important !!! ============
+When running the tape tests, Bacula will write on any tape that
+is in the tape drive that you have configured. If it is a production
+Bacula tape, it will be destroyed. If you have configured an Autochanger,
+Bacula will write on the tapes in slots 1 and 2 thus destroying any
+information on those tapes, even if they are Bacula production tapes.
+===================================
+
+Each of the above calls one or more scripts. By looking at the
+scripts available in this directory, you can see that there are a number
+of options for running tests.
You can run them individually as:
To avoid re-doing a make setup if you have made a change to the
conf files, and you do not need a new copy of the source, you can simply do:
- scripts/do-sed <your-email-address>
+ make sed
Debugging failed tests:
-The simplest thing to do is to edit tests/xxxx where xxxx is the name of
-a or the test that is failing and change all:
-
- @output
-
-to
-
- @tee
-
-re-run the test. The output will display on your terminal. In rare cases,
-for example to get bacula debug output, you might want to remove the
-"2>&1 >/dev/null" from the end of the bin/bacula call or from the end
-of the diff call, assuming you want to "see" the diff output.
-
+Prior versions required editing the tests/xxxx and changing a debug flag.
+However, that has been replaced by two environment variables:
+
+ REGRESS_DEBUG
+ REGRESS_WAIT
+
+If you define REGRESS_DEBUG, e.g.
+
+ REGRESS_DEBUG=1
+ export REGRESS_DEBUG
+
+then run a test, it will display the job and debug output.
+
+If you define REGRESS_WAIT, the script will stop and request:
+
+Start Bacula under debugger and enter anything when ready ...
+
+At this point, you can start any of the daemons under the debugger,
+then answer the message by entering any character. The script will
+then continue. For any daemon or daemons that you have manually started,
+you will see an error message when the script attempts to run a second
+copy, but those messages can be ignored. This makes it reasonably easy
+to run any component or components under the debugger if necessary.
+
+Explicit example:
+
+In shell window 1.
+
+cd regress
+export REGRESS_DEBUG=1
+export REGRESS_WAIT=1
+tests/name-of-script-test
+(wait until it tells you to start the debugger)
+
+In shell window 2
+
+cd regress/bin
+gdb bacula-xx (where xx is the component you want to debug).
+(possibly set a break point -- normally not)
+run -s -f
+(wait for the output to stop)
+
+In shell window 1
+(enter any character or simply a return)
+(ignore the error message it prints complaining that the daemon
+you are debugging is already running, which is in fact the case).
+
+
+That is all there is to it. The debugger window will get some
+output and will stop waiting for input if anything goes wrong
+like a seg fault. At that point, you can enter commands.
+
+The procedure avoids modifying the test scripts and trying to
+find pids and the such. If you want less debug output when
+debugging, don't set REGRESS_DEBUG=1.
+
+===
+
+Also, if you run from time to time on a computer that is not connected
+to the network, please be sure that "hostname" is set to "localhost",
+otherwise, your tests may fail because the hostname used by Bacula's
+./configure cannot be properly resolved.
+
+Anyway, you can debug where it is happening in the source code using the
+following example. For example, here I get the following backtrace:
+
+======= Backtrace: =========
+/lib/libc.so.6[0xb7b9d6e1]
+/lib/libc.so.6(cfree+0x89)[0xb7b9ed79]
+/home/kern/bacula/regress/bin/bacula-fd[0x8082ae5]
+/home/kern/bacula/regress/bin/bacula-fd[0x8082d58]
+/home/kern/bacula/regress/bin/bacula-fd[0x80838ac]
+/home/kern/bacula/regress/bin/bacula-fd[0x807aa3f]
+/home/kern/bacula/regress/bin/bacula-fd[0x807ac29]
+/home/kern/bacula/regress/bin/bacula-fd[0x804d188]
+/lib/libc.so.6(__libc_start_main+0xdc)[0xb7b4ef9c]
+/home/kern/bacula/regress/bin/bacula-fd[0x804cd21]
+
+Now to convert this into something more meaningful, kill off any hung Bacula
+processes. Note the one that was running -- above you see that it was
+bacula-fd, then bring the same binary up in the debugger. Then start at the
+first bacula-fd line, and feed the hex number to gdb as follows:
+
+info symbol 0x8082ae5
+free_addresses(dlist*) + 53 in section .text
+
+info symbol 0x8082d58
+add_address(dlist**, IPADDR::i_type, unsigned short, int, char const*, char
+const*, char**) + 568 in section .text
projects / bacula / bacula / blobdiff