== INSTALL == A.1 Linux Installation (rpm, deb) .2 Mac OS X Installation .3 AIX Installation .4 Solaris Installation .5 Windows Installation B. Building from source C. Post-install steps D. Non-default build, install and run This document describes how to configure and build the open source PCP ("pcp") package from source, and how to install and finally run it. == A.1 Linux Installation If you are using Debian, or a Debian-based distribution like Ubuntu, PCP is included in the distribution (as of late 2008). Run: # apt-get install pcp If you are using a RPM based distribution and have the binary rpm: # rpm -Uvh pcp-*.rpm ... and skip to the final section (below) - "Post-install steps". Special note for Ubuntu 8.04, 8.10, 9.04, 9.10 and 10.04 I've had to make the changes below to /usr/bin/dpkg-buildpackage. Without these two changes, my pcp builds produce bad binaries with a bizarre array of failure modes! #kenj# my $default_flags = defined $build_opts->{noopt} ? "-g -O0" : "-g -O2"; my $default_flags = defined $build_opts->{noopt} ? "-g -O0" : "-g -O0"; my %flags = ( CPPFLAGS => '', CFLAGS => $default_flags, CXXFLAGS => $default_flags, FFLAGS => $default_flags, #kenj# LDFLAGS => '-Wl,-Bsymbolic-functions', LDFLAGS => '', ); Without these changes, we see QA failures for 039, 061, 072, 091, 135, 147 and 151 ... and the QA 166 goes into a loop until it fills up the root filesystem. -- Ken == A.2 Mac OS X Installation The only unusual step in installing PCP on Mac OS X is the additional dependency on the developer tools (XCode package). The specific need PCP has is to have a C preprocessor available for managing the metric namespace, which on Mac OS X is a frontend to gcc and not installed by default. Once this is installed, installing PCP from the DMG file is as simple as clicking on the icon in a Finder window, and following the prompts from the installer. == A.3 AIX Installation At this stage, noone is making available pre-built AIX packages. A port to AIX has been done, and merged, however - building from the source is currently the only option. The packaging work is also begun on this platform (see the build/aix/ directory in the sources). == A.4 Solaris Installation Prebuild Solaris packages are available from PCP download section. At this stage the package are distributed as SVR4 package datastream and are build on Open Solaris 5.11. You can install the package using 'pkgadd' command, e.g.: # pkgadd -d pcp-X.Y.Z During the installation the following three services are registered with the Solaris' service management facility: # svccfg list \*/pcp/\* application/pcp/pmcd application/pcp/pmlogger application/pcp/pmie application/pcp/pmproxy On the new installation all services are disabled, during the upgrade from the previous version of PCP the state of the services is preserved. Use of 'svcadm' command to enable or disable is preferred over explicit invocation of the pmcd start script. Use 'svcs' command to check the state of the services, e.g.: # svcs -l application/pcp/pcp fmri svc:/application/pcp/pcp:default name Performance Co-Pilot Collector Daemon enabled false state disabled next_state none state_time 20 March 2012 11:33:27 AM EST restarter svc:/system/svc/restarter:default dependency require_all/none svc:/system/filesystem/local:default (online) svc:/milestone/network:default (online) == A.5 Windows Installation Download the native Windows version of PCP from the PCP download section. There are two variants: .zip and .exe. The latter is a self-installing executable, the former a simple compressed PCP image. Run the executable, follow the prompts, and a Startup Menu item with several PCP options will be available - standard DOS shell or POSIX shell (with suitable environment setup), the Perl CPAN configuration tool, links to documentation and online PCP internet resources, and the pmchart utility. == B. Building from source 1. Configure, build and install the package The pcp package uses autoconf/configure and expects a GNU build environment (your platform must at least have gmake). If you just want to spin a .RPM, .DEB, .DMG, .EXE and/or tar file, use the Makepkgs script in the top level directory. This will configure and build the package for your platform and leave binary and src packages in the build/ directory. It will also leave binary and source tar file in the build/tar directory. $ ./Makepkgs --verbose 2. If you want to build the package and install it manually you will first need to ensure the "user" pcp is created so that key parts of the PCP installation can run as a user other than root. For Debian this means the following (equivalent commands are available on all distributions): $ su root # groupadd -r pcp # useradd -c "Performance Co-Pilot" -g pcp -d /var/lib/pcp -M -r -s /usr/sbin/nologin pcp Then use the following steps (use configure options to suit your preferences, refer to the qa/admin/myconfigure script for some guidance and see also section D below for additional details): $ ./configure --prefix=/usr --libexecdir=/usr/lib --sysconfdir=/etc \ --localstatedir=/var --with-rcdir=/etc/init.d $ make $ su root # make install Note 0: PCP services run as non-root by default. Create unprivileged users "pcp" with home directory /var/lib/pcp, and "pcpqa" with home directory such as /var/lib/pcp/testsuite, or as appropriate, or designate other userids in the pcp.conf file. Note 1: that there are so many "install" variants out there that we wrote our own script (see "install-sh" in the top level directory), which works on every platform supported by PCP. Note 2: the Windows build is particularly involved to setup, this is primarily due to build tools not being available by default on that platform. See the PCP Glider scripts and notes in the pcpweb tree to configure your environment before attempting to build from source under Win32. == C. Post-install steps You will need to start the PCP Collection Daemon (PMCD), as root: Linux, AIX: # service pmcd start (or...) # /etc/init.d/pmcd start (or...) # /etc/rc.d/init.d/pmcd start Mac OS X: # /Library/StartupItems/pcp/pmcd start Windows: $PCP_DIR/etc/pmcd start Solaris: # svcadm enable application/pcp/pmcd Once you have started the PMCD daemon, you can list all performance metrics using the pminfo(1) command, E.g. # pminfo -fmdt (you don't have to be root for this, but you may need to type rehash so your shell finds the pminfo command). If you are writing scripts, you may find the output from pmprobe(1) easier to parse than that for pminfo(1). There are numerous other PCP client tools included. PCP can be configured to automatically log certain performance metrics for one or more hosts. The scripts to do this are documented in pmlogger_check(1). By default this facility is not enabled. If you want to use it, you need to # determine which metrics to log and how often you need them # edit $PCP_SYSCONF_DIR/pmlogger/control # edit $PCP_SYSCONF_DIR/pmlogger/config.default # (and any others in same dir) # as root, "crontab -e" and add something like: # -- typical PCP log management crontab entries # daily processing of pmlogger archives and pmie logs 10 0 * * * $PCP_BINADM_DIR/pmlogger_daily 15 0 * * * $PCP_BINADM_DIR/pmie_daily # # every 30 minutes, check pmlogger and pmie instances are running 25,40 * * * * $PCP_BINADM_DIR/pmlogger_check 5,55 * * * * $PCP_BINADM_DIR/pmie_check The pmie (Performance Metrics Inference Engine) daemon is _not_ configured to start by default. To enable it, you may want to (on Linux platforms with chkconfig). # su root # chkconfig pmie on # edit the pmie control file (usually below $PCP_SYSCONF_DIR/pmie) # edit the config file (usually $PCP_SYSCONF_DIR/pmie/config.default) # set up cron scripts similar to those for pmlogger (see above) Configure some optional Performance Metrics Domain Agents (PMDAs) The default installation gives you the metrics for cpu, per-process, file system, swap, network, disk, memory, interrupts, nfs/rpc and others. These metrics are handled using the platform PMDA - namely pmda_linux.so (Linux), pmda_darwin.dylib (Mac), or pmda_windows.dll (Windows). It also gives you the PMCD PMDA, which contains metrics that monitor PCP itself. There are many other optional PMDAs that you can configure, depending on which performance metrics you need to monitor, as follows: Note: $PCP_PMDAS_DIR is normally /var/pcp/pmdas, see pcp.conf(4). Web Server metrics # su root # cd $PCP_PMDAS_DIR/apache (i.e. cd /var/pcp/pmdas/apache) # ./Install # Check everything is working OK # pminfo -fmdt apache Other PMDAs in the pcp package include: apache - monitor apache web server stats cisco - monitor Cisco router stats dbping - query any database, extract response times elasticsearch - monitor an elasticsearch cluster kvm - monitor kernel-based virtual machine stats mailq - monitor the mail queue memcache - monitor memcache server stats mmv - export memory-mapped value stats from an application mounts - keep track of mounted file systems mysql - monitor MySQL relational databases postgres - monitor PostGreSQL relational databases process - keep an eye on critical processes/daemons roomtemp - monitor room temp (needs suitable probe) rsyslog - monitor the reliable system log daemon sendmail - monitor sendmail statistics shping - ping critical system services, extract response times trace - for instrumenting arbitrary applications, see pmtrace(1) txmon - transaction and QOS monitoring sample - for testing simple - example src code if you want to write a new PMDA trivial - even easier src code for a new PMDA. The procedure for configuring all of these is to change to the directory for the PMDA (usually below /var/lib/pcp/pmdas), and then run the ./Install script found therein. None of these PMDAs are configured by default - you choose the PMDAs you need and run the Install script. Installation can be automated (defaults chosen) by touching .NeedInstall in the appropriate pmdas directory and then restarting the pmcd service via its startup script. == D. Non-default build, install and run To run build and run a version of PCP that is installed in a private location (and does not require root privileges), first create the pcp "user" as described in section B.2 above), then $ ./configure --prefix=/some/path This will populate /some/path with a full PCP installation. To use this ensure the following are set in the environment: $ export PCP_DIR=/some/path Amend your shell's $PATH to include the PCP directories, found as follows: $ cd /some/path $ xtra=`grep '^PCP_BIN' etc/pcp.conf | sed -e 's/.*=//' | paste -s -d :` $ PATH=$xtra:$PATH Ensure the new libraries can be found: $ export LD_LIBRARY_PATH=`grep '^PCP_LIB' etc/pcp.conf \ | sed -e 's/.*=//' | uniq | paste -s -d :` Tell Perl where to find loadable modules: $ export PERL5LIB=$PCP_DIR/usr/lib/perl5:$PCP_DIR/usr/share/perl5 Allow man(1) to find the PCP manual pages: $ export MANPATH=`manpath`:$PCP_DIR/usr/share/man If your version is co-exiting with a running PCP in a default install, then alternative port numbers in your environment for pmcd ($PMCD_PORT), pmlogger ($PMLOGGER_PORT) and pmproxy ($PMPROXY_PORT)