diff options
Diffstat (limited to '')
| -rw-r--r-- | sql-bench/Comments/Access.crash-me | 40 | ||||
| -rw-r--r-- | sql-bench/Comments/Adabas.crash-me | 36 | ||||
| -rw-r--r-- | sql-bench/Comments/Empress.crash-me | 102 | ||||
| -rw-r--r-- | sql-bench/Comments/FrontBase.benchmark | 59 | ||||
| -rw-r--r-- | sql-bench/Comments/Informix.crash-me | 26 | ||||
| -rw-r--r-- | sql-bench/Comments/interbase | 18 | ||||
| -rw-r--r-- | sql-bench/Comments/mysql.benchmark | 39 | ||||
| -rw-r--r-- | sql-bench/Comments/postgres.benchmark | 107 | ||||
| -rw-r--r-- | sql-bench/Comments/postgres.crash-me | 30 |
9 files changed, 457 insertions, 0 deletions
diff --git a/sql-bench/Comments/Access.crash-me b/sql-bench/Comments/Access.crash-me new file mode 100644 index 00000000..f4a419aa --- /dev/null +++ b/sql-bench/Comments/Access.crash-me @@ -0,0 +1,40 @@ +Access 97 tested through ODBC 1998.04.19, by monty@mysql.com + +Access 97 has a bug when on executes a SELECT follwed very fast with a +DROP TABLE or a DROP INDEX command: + +[Microsoft][ODBC Microsoft Access 97 Driver] The database engine couldn't lock table 'crash_q' because it's already in use by another person or process. (SQL-S1 +000)(DBD: st_execute/SQLExecute err=-1) + +Debugging SQL queries in Access 97 is terrible because most error messages +are of type: + +Error: [Microsoft][ODBC Microsoft Access 97 Driver] Syntax error in CREATE TABLE statement. (SQL-37000)(DBD: st_prepare/SQLPrepare err=-1) + +Which doesn't tell a thing! + +-------------- + +Access 2000 tested through ODBC 2000.01.02, by monty@mysql.com + +crash-me takes a LONG time to run under Access 2000. + +The '1+NULL' and the 'OR and AND in WHERE' tests kills +Activestate Perl, build 521, DBI-DBC with an OUT OF MEMORY error. +The later test also kills perl/access with some internal errors. +To go around this one must run crash-me repeatedly with the --restart option. + +Testing of the 'constant string size' (< 500K) takes a LOT of memory +in Access (at least 250M on My computer). + +Testing of number of 'simple expressions' takes REALLY a lot of time +and memory; At some point I was up to 350M of used memory! + +To fix the above, I modified crash-me to have lower max limits in the +above tests. + +Benchmarks (under Win98): + +Running the connect-test will take up all available memory and this +will not be freed even after quitting perl! There is probably some +bug in the Access connect code that eats memory! diff --git a/sql-bench/Comments/Adabas.crash-me b/sql-bench/Comments/Adabas.crash-me new file mode 100644 index 00000000..d36d0504 --- /dev/null +++ b/sql-bench/Comments/Adabas.crash-me @@ -0,0 +1,36 @@ + +I did not spend much time for tuning crash-me or the limits file. In short, +here's what I did: + + - Put engine into ANSI SQL mode by using the following odbc.ini: + + [ODBC Data Sources] + test + + [test] + ServerDB=test + ServerNode= + SQLMode=3 + + - Grabbed the db_Oracle package and copied it to db_Adabas + - Implemented a 'version' method. + - Ran crash-me with the --restart option; it failed when guessing the + query_size. + - Reran crash-me 3 or 4 times until it succeeded. At some point it + justified its name; I had to restart the Adabas server in the + table name length test ... + - Finally crash-me succeeded. + +That's it, folks. The benchmarks have been running on my P90 machine, +32 MB RAM, with Red Hat Linux 5.0 (Kernel 2.0.33, glibc-2.0.7-6). +Mysql was version 3.21.30, Adabas was version 6.1.15.42 (the one from +the promotion CD of 1997). I was using X11 and Emacs while benchmarking. + +An interesting note: The mysql server had 4 processes, the three usual +ones and a process for serving me, each about 2 MB RAM, including a +shared memory segment of about 900K. Adabas had 10 processes running from +the start, each about 16-20 MB, including a shared segment of 1-5 MB. You +guess which one I prefer ... :-) + + +Jochen Wiedmann, joe@ispsoft.de diff --git a/sql-bench/Comments/Empress.crash-me b/sql-bench/Comments/Empress.crash-me new file mode 100644 index 00000000..b60bf4f1 --- /dev/null +++ b/sql-bench/Comments/Empress.crash-me @@ -0,0 +1,102 @@ +***************************************************************** +NOTE: +This is an old comment about how it was to run crash-me on empress +the first time. I think it was on Empress 6.0 +***************************************************************** + +start testing empress ... +added a nice line for the max join .... +strip the as out of the from field ... +that's working on empress .... + +at this moment with .... +max constant string size in where .... taking a lot of memory ... +at this moment (it's still growing just waiting till it stops ..) 99mb .. +sorry it started growing again ... +max 170 mb ... then it gives an error ... +Yes it crashed ..... +at max constant string size in where ... with IOT trap/Abort(core dumped) :-) +nice isn't it ... hope it saved the things .... +I outcommented the sig story because I could see how the script is running +and I wasn't sure if SIG{PIPE} ='DEFAULT' ... is working ... +restarting with limit 8333xxx ... couldn't see it any more ... +query is printed ...(200000 lines ..). mmm Nice IOT trap/Abort ... +and again ..and again ... +aha ... and now it's going further ... +max constant string string size in select: ... +taking 100 mb +crashing over and over again .... +max simple expressions ... +is taking ... 82 mb ... +mmmm this is taking very very very long .... after 10 minutes I will kill it and run it again ... I think he can't proces this query that fast ... and will crash any way ... +still growing very slow to the 90 mb ... +killed it ... strange is ... it don't react on ctrl-c ... but kill 15 does work +mmm still bussy with killing his self ... memory is growing to 128 mb ... +sorry .. 150 mb .. and then the output .. +maybe something for the extra things for crash-me ... +if debug .... +if length $query > 300 ... just print $errstr .. else print $query + $errstr .. +at this moment he is still bussy printing .... +first clear all locks ... with empadm test lockclear ... else it will give me +the error with a lock ... +restarting at 4194297 .... mmm a bit high I think ... +after 5 minutes I will kill it ... +mmm have to kill it again ... took 30 mb ..now growing to 42 mb .. +restarting at 838859 ... hope this will crash normaly ... :-) +I will give it again 5 minutes to complete ... +taking 12 mb .... will kill it ... after 4 minutes .... +restarting at 167771 ... taking 6 mb ... give it again 5 minutes .... + will kill it again ... else it becomes to late tonight ... +mmm started with 33xxxx and it crashes ...:-) yes ... +can't we build in a function which will restart his self again ... +mmmm this is really boring .. start it over and over again ... +WHO .... NICE >>>> +Restarting this with high limit: 4097 +................. +*** Program Bug *** setexpr: unknown EXPR = 1254 (4e6) +isn't it ... starting it again ... +finally finished with 4092 .... +now max big expression ..... +directly taking .. 85 mb ... give it again 5 minutes ... +mmm I am going to kill it again ... mmm it grows to 146 mb ... +restarting with 1026 ... taking 25 mb .. +won't give him that long ... because it will crash any way (just a ques) .. +killed it ... +restarting at 205 ... hope this will work .... +won't think so ... give it 2 minutes ... taking 12 mb ... +killed it ...restarting at ... 40 ... yes it crashes ... + 7 is crashing ... 1 ....is good .. finaly ... a long way ... +now max stacked expressions .... +taking 80 mb ... mmmm what sort of test is this ...it looks more like a harddisk test .. but it crashes .. nice ... +mmm a YACC overflow ... that's a nice error ... +but it goes on ... yep it didn't crashed just an error ... + mmm +my patch for the join didn't work ... let's take a look what goes wrong ... +saw it ... forgot some little thing .. mm not .. them ... another little typo +mmm again a really nice bug ... +Restarting this with high limit: 131 +... +*** Program Bug *** xflkadd: too many read locks +them the lock forgotten .... +mmmm bigger problem ... +with empadm test lockinfo ... gives ... +*** System Problem *** no more clients can be registered in coordinator + +*** User Error *** '/usr/local/empress/rdbms/bin/test' is not a valid database +that's really really nice .... +hmmm after coordclear ... it's fine again ... +strange ... + after restarting it again the script ... it is going further .... +the overflow trick is nice and working good ... +now I have table 'crash_q' does not exist for every thing ... +normal ...???? mmm went after all good .. so I think it's normal ... +mmmm a lot of table 'crash_q' does not exist ... again ... +sometimes when the overflow is there ... I restart it and it is saying ... +restarting at xxxx that's not good ... but hey ... what the hack ... +maybe that's good because if one test run's more then 200 times .... +it won't exceeds that test ... +.... +yes finally the end of crash-me ... +at last ... crash-me safe: yes ... +yep don't think so he .... + diff --git a/sql-bench/Comments/FrontBase.benchmark b/sql-bench/Comments/FrontBase.benchmark new file mode 100644 index 00000000..03386a4d --- /dev/null +++ b/sql-bench/Comments/FrontBase.benchmark @@ -0,0 +1,59 @@ +# This file describes how to run benchmarks and crash-me with FrontBase + +Installed components: + +- FrontBase-2.1-8.rpm + (had to run with rpm -i --nodeps; the rpm wanted libreadline.so.4.0, + but only libreadline.so.4.1 was available) + +- DBD-FB-0.03.tar.gz + (perl Makefile.Pl; + make; + make test; + make install;) + +- DBI-1.14.tar.gz + (perl Makefile.Pl; + make; + make test; + make install;) + +- Msql-Mysql-modules-1.2215.tar.gz + (perl Makefile.Pl; + make; + make test; + make install;) + +After installations: + +- cd /etc/rc.d + FBWeb start + FrontBase start + +- cd /usr/local/mysql/sql-bench +- FBExec & +- FrontBase test + +crash-me: + +There were a lot of troubles running the crash-me; FrontBase core +dumped several tens of times while crash-me was trying to determine +the maximum values in different areas. + +The crash-me program itself was also needed to be tuned quite a lot +for FB. There were also some bugs/lacking features in the crash-me +program, which are now fixed to the new version. + +After we finally got the limits, we runned the benchmarks. + +benchmarks: + +Problems again. Frontbase core dumped with every part of the +benchmark (8/8) tests. After a lot of fine-tuning we got the +benchmarks to run through. The maximum values had to be dropped +down a lot in many of the tests. + +The benchmarks were run with the following command: + +perl run-all-tests --server=frontbase --host=prima +--cmp=frontbase,mysql --tcpip --log diff --git a/sql-bench/Comments/Informix.crash-me b/sql-bench/Comments/Informix.crash-me new file mode 100644 index 00000000..557db012 --- /dev/null +++ b/sql-bench/Comments/Informix.crash-me @@ -0,0 +1,26 @@ +***************************************************************** +NOTE: +I, Monty, pulled this comment out from the public mail I got from +Honza when he published the first crash-me run on Informix +***************************************************************** + +Also attached are diffs from server-cfg and crash-me -- some of +them are actual bugs in the code, some add extensions for Informix, +some of the comment-outs were necessary to finish the test. Some of +the problematic pieces that are commented out sent Informix to +veeeery long load 1 on the machine (max_conditions for example), so +could be considered crashes, but I'd prefer that someone checks the +code before giving out such a conclusion. + +Some of the code that is commented out failed with some other SQL +error message which might mean a problem with the sequence of commands +in crash-me. Interesting thing, some of the tests failed for the +first time but in the next or third run went OK, so the results are +results of more iterations (like column doesn't exist in the first +try but the second pass goes OK). + +I'd like to hear your comments on the bug fixes and Informix specific +code before we go into debugging the problems. + +Yours, + Honza Pazdziora diff --git a/sql-bench/Comments/interbase b/sql-bench/Comments/interbase new file mode 100644 index 00000000..addaf74b --- /dev/null +++ b/sql-bench/Comments/interbase @@ -0,0 +1,18 @@ +Running crash-me on Interbase: +I +- got opensource version of interbase 6.0.1 + (both mode, classic and superserver), +- set up DBD::InterBase from cpan, +- created database "test" and set sql_dialect for that database to 3 +- executed crash-me for both interbase's models (classic and superserver). + +There were some problems during the execution: +1) Sometimes client side got SIGSEGV , At that moment server side + writes + gds__alloc: non-positive size allocation request + to log file. + This problem has both models. I am not shure if it's interbase or + DBD:InterBase problem (though DBD::InterBase made all nesessary + tests during the installation without any problem) + +2) In "superserver" mode ibserver several times died (and ibguard restarted it) diff --git a/sql-bench/Comments/mysql.benchmark b/sql-bench/Comments/mysql.benchmark new file mode 100644 index 00000000..628903de --- /dev/null +++ b/sql-bench/Comments/mysql.benchmark @@ -0,0 +1,39 @@ +# This file describes how to run MySQL benchmarks with MySQL +# + +# The test was run on a Intel Xeon 2x 550 Mzh machine with 1G memory, +# 9G hard disk. The OS is Suse 6.4, with Linux 2.2.14 compiled with SMP +# support +# Both the perl client and the database server is run +# on the same machine. No other cpu intensive process was used during +# the benchmark. + +# +# +# First, install MySQL from RPM or compile it according to the +# recommendations in the MySQL manual +# + +# Start MySQL + +bin/safe_mysqld --key_buffer=16M & + +# +# Now we run the test that can be found in the sql-bench directory in the +# MySQL 3.23 source distribution with and without --fast +# +# Note that if you want to make a results that is comparead to some database, +# You should add "--cmp=databasename" as an extra option to the test +# +$CMP=--cmp=pg + +run-all-tests --comment="Intel Xeon, 2x550 Mhz, 1G, key_buffer=16M" $CMP +run-all-tests --comment="Intel Xeon, 2x550 Mhz, 1G, key_buffer=16M" --fast $CMP + +# If you want to store the results in a output/RUN-xxx file, you should +# repeate the benchmark with the extra option --log --use-old-result +# This will create a the RUN file based of the previous results +# + +run-all-tests --comment="Intel Xeon, 2x550 Mhz, 1G, key_buffer=16M" --log --use-old-result $CMP +run-all-tests --comment="Intel Xeon, 2x550 Mhz, 1G, key_buffer=16M" --fast --log --use-old-result $CMP diff --git a/sql-bench/Comments/postgres.benchmark b/sql-bench/Comments/postgres.benchmark new file mode 100644 index 00000000..c52a5369 --- /dev/null +++ b/sql-bench/Comments/postgres.benchmark @@ -0,0 +1,107 @@ +# This file describes how to run MySQL benchmark suite with PostgreSQL +# +# WARNING: +# +# Don't run the --fast test on a PostgreSQL 7.1.1 database on +# which you have any critical data; During one of our test runs +# PostgreSQL got a corrupted database and all data was destroyed! +# When we tried to restart postmaster, It died with a +# 'no such file or directory' error and never recovered from that! +# +# Another time vacuum() filled our system disk with had 6G free +# while vaccuming a table of 60 M. +# +# WARNING + +# The test was run on a Intel Xeon 2x 550 Mzh machine with 1G memory, +# 9G hard disk. The OS is Suse 7.1, with Linux 2.4.2 compiled with SMP +# support +# Both the perl client and the database server is run +# on the same machine. No other cpu intensive process was used during +# the benchmark. +# +# During the test we run PostgreSQL with -o -F, not async mode (not ACID safe) +# because when we started postmaster without -o -F, PostgreSQL log files +# filled up a 9G disk until postmaster crashed. +# We did however notice that with -o -F, PostgreSQL was a magnitude slower +# than when not using -o -F. + +# +# First, install postgresql-7.1.2.tar.gz + +# Adding the following lines to your ~/.bash_profile or +# corresponding file. If you are using csh, use īsetenvī. + +export POSTGRES_INCLUDE=/usr/local/pg/include +export POSTGRES_LIB=/usr/local/pg/lib + +PATH=$PATH:/usr/local/pg/bin +MANPATH=$MANPATH:/usr/local/pg/man + +# +# Add the following line to /etc/ld.so.conf: +# + +/usr/local/pg/lib + +# and run: + +ldconfig + +# untar the postgres source distribution, cd to postgresql-* +# and run the following commands: + +CFLAGS=-O3 ./configure +gmake +gmake install + +mkdir /usr/local/pg/data +chown postgres /usr/local/pg/data +su - postgres +/usr/local/pg/bin/initdb -D /usr/local/pg/data +/usr/local/pg/bin/postmaster -o -F -D /usr/local/pg/data & +/usr/local/pg/bin/createdb test +exit + +# +# Second, install packages DBD-Pg-1.00.tar.gz and DBI-1.18.tar.gz, +# available from http://www.perl.com/CPAN/ + +export POSTGRES_LIB=/usr/local/pg/lib/ +export POSTGRES_INCLUDE=/usr/local/pg/include/postgresql +perl Makefile.PL +make +make install + +# +# Now we run the test that can be found in the sql-bench directory in the +# MySQL 3.23 source distribution. +# +# We did run two tests: +# The standard test + +run-all-tests --comment="Intel Xeon, 2x550 Mhz, 512M, pg started with -o -F" --user=postgres --server=pg --cmp=mysql + +# When running with --fast we run the following vacuum commands on +# the database between each major update of the tables: +# vacuum anlyze table +# vacuum table +# or +# vacuum analyze +# vacuum + +# The time for vacuum() is accounted for in the book-keeping() column, not +# in the test that updates the database. + +run-all-tests --comment="Intel Xeon, 2x550 Mhz, 512M, pg started with -o -F" --user=postgres --server=pg --cmp=mysql --fast + +# If you want to store the results in a output/RUN-xxx file, you should +# repeate the benchmark with the extra option --log --use-old-result +# This will create a the RUN file based of the previous results + +run-all-tests --comment="Intel Xeon, 2x550 Mhz, 512M, pg started with -o -F" --user=postgres --server=pg --cmp=mysql --log --use-old-result +run-all-tests --comment="Intel Xeon, 2x550 Mhz, 512MG, pg started with -o -F" --user=postgres --server=pg --cmp=mysql --fast --log --use-old-result + +# Between running the different tests we dropped and recreated the PostgreSQL +# database to ensure that PostgreSQL should get a clean start, +# independent of the previous runs. diff --git a/sql-bench/Comments/postgres.crash-me b/sql-bench/Comments/postgres.crash-me new file mode 100644 index 00000000..c693817c --- /dev/null +++ b/sql-bench/Comments/postgres.crash-me @@ -0,0 +1,30 @@ +***************************************************************** +NOTE: +This is an old comment about how it was to run crash-me on postgreSQL +the first time. I think it was on pg 6.2 +***************************************************************** + +mmm memory use of postgres is very very much ... +at this moment I am testing it ... +and the tables in join: is taking 200MB memory ... +I am happy to have 400mb swap ... so he can do have it ... +but other programs will give some errors ... +just a second ago ... vim core dumped .. XFree crashed full ... to the prompt +the menu bar of redhat disappeared .... +at this momemt the max is 215 mb memore postgres is taking ... + +the problem with postgres is the following error: +PQexec() -- Request was sent to backend, but backend closed the channel before r +esponding. This probably means the backend terminated abnormally before or whil +e processing the request + +I think we can solve this with a goto command ... to go back again ... after +the connect again ... +postgres is taking 377 mb .... mmm allmost out of memory ... 53mb left .. +mmm it's growing ... 389 mb ..393 mb ... 397 mb .. better can wait for the out of memory ... i think 409 412 max ... + +ps added some nice code for the channel closing ... +it must now do again the query when the error is the above error ... +hopes this helps ... +after crashing my X again ... +I stopped testing postgres |