Hi Serge, Thanks!!
I confirmed your patch in postgresql-8.4.4 on RHEL5.5. Of course it combined Cluster-Resource-Agents-784385d01d0f.tar and confirmed it. There were some problems to your patch. I revised it as follows. runasowner() { local quietrun="" [ "x$1" = "x-q" ] && { quietrun="-q" shift 1 -------> need shift!! } ocf_run $quietrun su $OCF_RESKEY_pgdba -c "$*" } The problem was solved with your patch. Let a development version please reflect a patch. I attach the first patch which I changed. Best Regards, Hideo Yamauchi. --- renayama19661...@ybb.ne.jp wrote: > Hi Serge, > > Thank you for a patch. > I test a patch and will report a result next week. > > > Hideo-san, I don't think that we need to introduce a new parameter to > > control verbosity of monitor. There is nothing really important in > > that output. > > All right. > I agree that I do not give the log of monitor. > > > Best Regards, > Hideo Yamauchi. > > > --- Serge Dubrouski <serge...@gmail.com> wrote: > > > Ok, attached are 2 patches for pgsql RA. First adds use for the new > > "-q" option of ocf_run. Second fixes "meta-data" method plus there is > > a rather deep analysis of exit code > > > > Hideo-san, I don't think that we need to introduce a new parameter to > > control verbosity of monitor. There is nothing really important in > > that output. > > > > On Fri, Aug 20, 2010 at 2:45 AM, Dejan Muhamedagic <deja...@fastmail.fm> > > wrote: > > > Hi, > > > > > > On Thu, Aug 19, 2010 at 08:43:38AM -0600, Serge Dubrouski wrote: > > >> Ahh, thanks. I've got where is the problem, and just wonder if > > >> apache > > >> RA suffers from the same issue. > > >> > > >> While it is possible to fix this in RA by not calling get_pgsql_param > > >> if the script was called with for "meta-data" (and in this case that > > >> will be impossible to provide correct default value for scoket_dir if > > >> user changed it in configuration file) I'm not sure if it's correct to > > >> not pass all of OCF_RESKEY parameters down to RA when "meta-data" > > >> function is called. Dejan, what do you think? Shall we create a > > >> Bugzilla for this issue and get it fixed in Pacemaker? > > > > > > The meta-data is most often invoked outside of pacemaker, for > > > instance by 'crm ra' or when the agents are packaged while > > > creating the man pages. It has to function without any > > > parameters. So, get_pgsql_param should not be called if the RA > > > was invoked to produce meta-data. > > > > > > As for defaults for various parameters, that's tricky, and should > > > be avoided if it depends on the content of an external file, > > > because a file can change at any time (also accidentally). > > > > > > Cheers, > > > > > > Dejan > > > > > >> On Thu, Aug 19, 2010 at 1:25 AM, <renayama19661...@ybb.ne.jp> > > >> wrote: > > >> > Hi Serge, > > >> > > > >> > Thank you for comment. > > >> > > > >> > My English interpretation may be wrong, but comments. > > >> > > > >> > $OCF_RESKEY_config is not handed to pgsql even if I set right > > >> > $OCF_RESKEY_config in the > > case of the > > >> > meta-data processing. > > >> > > > >> > static char* > > >> > get_resource_meta(const char* rsc_type, const char* provider) > > >> > { > > >> > const int BUFF_LEN=4096; > > >> > int read_len = 0; > > >> > char buff[BUFF_LEN]; > > >> > char* data = NULL; > > >> > GString* g_str_tmp = NULL; > > >> > char ra_pathname[RA_MAX_NAME_LENGTH]; > > >> > FILE* file = NULL; > > >> > GHashTable * tmp_for_setenv; > > >> > struct timespec short_sleep = > > >> > {0,200000000L}; /*20ms*/ > > >> > > > >> > get_ra_pathname(RA_PATH, rsc_type, > > >> > provider, ra_pathname); > > >> > > > >> > strncat(ra_pathname, " > meta-data",RA_MAX_NAME_LENGTH-strlen(ra_pathname)-1); > > >> > tmp_for_setenv = > > >> > g_hash_table_new(g_str_hash, g_str_equal); > > >> > add_OCF_env_vars(tmp_for_setenv, > > >> > "DUMMY_INSTANCE", rsc_type, > provider); -> **** > > setenv Only > > >> > environment vars. ********** > > >> > > > >> > raexec_setenv(tmp_for_setenv); > > >> > g_hash_table_foreach_remove(tmp_for_setenv, > let_remove_eachitem, > NULL); > > >> > g_hash_table_destroy(tmp_for_setenv); > > >> > (snip) > > >> > > > >> > In addition, when we use logd, this error is given, but it is thrown > > >> > away in ha_log. > > >> > > > >> >> The other option would be to change pgsql RA in a way that it won't > > >> >> parse config file if $OCF_RESKEY_config isn't set and in this case > > >> >> assume that $OCF_RESKEY_socketdir isn't set as well. That could > > >> >> potentially create problems for Ubuntu users for whom that > > >> >> functionality was created. > > >> > > > >> > Yes. > > >> > I know that there is a function of current pgsql for other users. > > >> > But I think that the analysis of config file is not necessary for > > >> > another many users. > > >> > In addition, in our environment, pgsql moves without analysis. > > >> > > > >> > Best Regards, > > >> > Hideo Yamauchi. > > >> > > > >> > > > >> > --- Serge Dubrouski <serge...@gmail.com> wrote: > > >> > > > >> >> Hello Hideo - > > >> >> > > >> >> I'm a bit lost with this info. If I understand you correct your > > >> >> postgresql.conf isn't in /var/lib/pgsql directory and pgsql RA script > > >> >> can't find it if $OCF_RESKEY_config isn't set. Then why not to just > > >> >> simply set $OCF_RESKEY_config and point it to correct place instead of > > >> >> introducing new parameter to disable parsing of configuration file? > > >> >> > > >> >> The other option would be to change pgsql RA in a way that it won't > > >> >> parse config file if $OCF_RESKEY_config isn't set and in this case > > >> >> assume that $OCF_RESKEY_socketdir isn't set as well. That could > > >> >> potentially create problems for Ubuntu users for whom that > > >> >> functionality was created. > > >> >> > > >> >> > > >> >> On Wed, Aug 18, 2010 at 8:46 PM, <renayama19661...@ybb.ne.jp> > > >> >> wrote: > > >> >> > Hi All, > > >> >> > > > >> >> > We confirmed that an error of pgsql occurred by the setting that we > > >> >> > did not use logd > > for. > > >> >> > When we use a syslog, this occurs. > > >> >> > And when $OCF_RESKEY_config is handled by default, an error occurs > > >> >> > when meta-data > > processing > > >> >> is > > >> >> > performed. > > >> >> > �* OCF_RESKEY_config of the user will not be > > >> >> > /var/lib/pgsql/data in many cases. > > >> >> > > > >> >> > (snip) > > >> >> > Aug 19 11:34:09 srv01 crmd: [9317]: info: match_graph_event: Action > > >> >> prmIpPostgreSQLDB2_monitor_10000 > > >> >> > (10) confirmed on srv01 (rc=0) > > >> >> > Aug 19 11:34:09 srv01 pgsql[9870]: INFO: could not change directory > > >> >> > to > > >> >> "/var/lib/heartbeat/cores/root" > > >> >> > server starting > > >> >> > Aug 19 11:34:09 srv01 pgsql[9870]: INFO: PostgreSQL start command > > >> >> > sent. > > >> >> > Aug 19 11:34:09 srv01 pgsql[9870]: INFO: PostgreSQL is down > > >> >> > Aug 19 11:34:10 srv01 pgsql[9870]: INFO: PostgreSQL is started. > > >> >> > Aug 19 11:34:10 srv01 pgsql[9975]: ERROR: Cannot find configuration > > >> >> > file > > >> >> > /var/lib/pgsql/data/postgresql.conf ----> ***** An error of the > > >> >> > meta-data processing > > after the > > >> >> start > > >> >> > processing > > >> >> > (snip) > > >> >> > Aug 19 11:35:09 srv01 pgsql[10526]: ERROR: Cannot find > > >> >> > configuration file > > >> >> > /var/lib/pgsql/data/postgresql.conf ----> ***** An error of the > > >> >> > �lrmadmin -M > ocf > > pgsql > > >> >> heartbeat > > >> >> > processing > > >> >> > (snip) > > >> >> > > > >> >> > Even if pgsql treats OCF_RESKEY_config by meta-data processing, > > >> >> > there is not a > meaning. > > >> >> > Because OCF_RESKEY_config is because it is not set. > > >> >> > > > >> >> > �* ha.cf > > >> >> > > > >> >> > pacemaker on > > >> >> > #use_logd on > > >> >> > debug 0 > > >> >> > udpport 694 > > >> >> > keepalive 2 > > >> >> > warntime 20 > > >> >> > deadtime 24 > > >> >> > initdead 48 > > >> >> > logfacility local1 > > >> >> > > > >> >> > The pgsql analyze a config file by get_pgsql_param processing. > > >> >> > There is the problem in this analysis. > > >> >> > > > >> >> > Two revisions think me to be thought. > > >> >> > > > >> >> > Pattern 1) A method to choose analysis of config file in an ocf > > >> >> > parameter. > > >> >> > �* For example, add a parameter of parse_config to pgsql. > > >> >> > �* Because many users do not need analysis of config file. > > >> >> > > > >> >> > Pattern 2) By a call except start,monitor,status,stop, > > >> >> > �pgsql do not do > > get_pgsql_param > > >> >> processing. > > >> >> > > > >> >> > It is the problem that I limited when cluster do not use logd, but > > >> >> > please please tell > an > > >> >> opinion of > > >> >> > all of you. > > >> >> > > === 以下のメッセージは省略されました ===
pgsql_patch1_0823
Description: 3450721922-pgsql_patch1_0823
_______________________________________________________ Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev Home Page: http://linux-ha.org/