mypapers/enterprise-software/enterprise-software.html

   1 <!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN">
   2 <html>
   3   <head>
   4     <link rel="stylesheet" href="../../mrtg-td/slides.css" type="text/css">
   5     <meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
   6     <meta name="Language" content="en">
   7     <meta name="Author" content="Petter Reinholdtsen">
   8   </head>
   9   <body>
  10
  11     <h1>Writing enterprise ready software</h1>
  12
  13     <p></p>
  14
  15     <p><tt><a href="http://www.hungry.com/~pere/mypapers/enterprise-software/enterprise-software.html">http://www.hungry.com/~pere/mypapers/enterprise-software/enterprise-software.html</a></tt></p>
  16
  17     <div class="presenter">Petter Reinholdtsen
  18       <br>pere@hungry.com
  19       <br>Debconf5, Helsinki 2005-06-12</div>
  20
  21     <h2>Overview</h2>
  22     <ul>
  23
  24      <li>we are in trouble
  25      <li>clues for the clueless
  26      <li>multilevel configuration
  27
  28    </ul>
  29
  30 <h2>We are in trouble</h2>
  31
  32 <p>With 60000 users and about 150 home directory file servers
  33  available from 12000 machines, users move from file server to file
  34  server.  This break several applications when the path to the users
  35  home directory changes.  (example: /mn/hegel/u1/pere to
  36  /usit/saruman/u1/pere).</p>
  37
  38 <!--
  39 Moving a user from one user disk to another break mozilla and
  40 openoffice configuration.  - not using relative paths
  41 -->
  42
  43 <h2>More trouble</h2>
  44
  45 <p>With 900 linux machines with common user database and home
  46 directories while running different versions of programs, users will
  47 run several versions of a program with the same configuration
  48 files.</p>
  49
  50 <!-- Trying to run with KDE 2 and 3 on different machines messes up
  51 the configuration - not handling downgrades
  52
  53 - handle up/downgrades - people will use different versions on
  54 different machines, with common home directories
  55
  56  -->
  57
  58 <h2>Some old trouble</h2>
  59
  60
  61 <p>Some users lost the source of their production systems, and need
  62 the binaries to keep working for 10-15 years.</p>
  63
  64 <!-- 15 year old binaries should keep working.  The source is lost
  65 long time ago, and the program is still used in production. -->
  66
  67 <h2>No room for more trouble</h2>
  68
  69 <p>When the file system for the PostgreSQL database in production goes
  70 full, one do not want to kick out 30000 users to take down the
  71 database and resize the file system.</p>
  72
  73 <h2>Disktrouble</h2>
  74
  75 <p>RAID is only only useful until the last redundancy disk is lost.
  76
  77 <p>Automatic RAID status systems need API or command line tools to
  78 extract the status.  Not like afacli, which go into interactive mode
  79 when an error is detected.
  80
  81 <h2>Installation trouble</h2>
  82
  83 <p>Trying to compile/install software on Irix, Solaris, Linux, HP-UX,
  84 Tru64 Unix, MacOSX and AIX when the process require a sysadmin to sit
  85 around to answer questions, change CDs, or insert licenses is both
  86 painful and prone to errors.</p>
  87
  88 <h2>Network trouble</h2>
  89
  90 <p>Trying to get some network server to work when it require to use
  91 some given port range, which is already taken by some other service
  92 and blocked in the router -- or try to get the coorporate network
  93 gatekeeper to open up the firewall</p>
  94
  95 <h2>Version trouble</h2>
  96
  97 <p>Given three tcl or php applications, is there one version of TCL or
  98 PHP usable with all of these?
  99
 100 <h2>Usability trouble</h2>
 101
 102 <p>When starting a program from the meny, where does it go if nothing
 103 appear on the screen?
 104
 105 <p>Do users always read their ~/.xsession-error file?
 106
 107
 108 Providing hooks to the local admins
 109
 110 backdoor - alternative way in.  fixing 800 machines by hand is not
 111 possible. - ssh config fuckup
 112
 113 reduce flexibility.  trying to support people over the phone when the
 114 gui is different for every person is a pain.
 115
 116
 117 <h2>Clues for the clueless</h2>
 118
 119 <ul>
 120
 121 <li>at least three levels of config files; package defaults, site
 122   defaults and host defaults
 123
 124 <li>never ask questions at compile time.  when compiling automatically
 125   for 10 platforms, a sysadmin do not want to sit down and answer
 126   questions.
 127
 128 <li>split installation tasks in two, one for installation, and one for
 129   the operations needing root access. (build / configuration as well)
 130
 131 <li>make sure the software can be installed anywhere (location
 132   independent), avoid hardcoding paths into the binaries.
 133
 134 <li>make the source available to make it possible to fix problems on
 135   site, and to use it on different platforms (os/hw) in the future
 136
 137 <li>Make paths into users home directories relative to ~user/, as users
 138   will move from disk to disk, or copy their home directory from site
 139   to site.  Always convert paths when saving config files.
 140
 141 <li>depend on as few libraries as possible, as it is a pain to get every
 142   extra library in place
 143
 144 <li>use well known libraries instead of making your own implementation.
 145   reduses the security risk.
 146
 147 <li>make sure libraries, and programming languages are backwards
 148   compatible.
 149
 150 <li>use a well known license.  it is a pain to evaluate every new
 151   license
 152
 153 <li>when distributing source, do not use vendor specific compiler
 154   features.  It will not work with the other vendors compilers used to
 155   compile on site.
 156
 157 <li>write portable code, make sure it works the same on all platforms.
 158
 159 <li>make the software work out of the box (require as little
 160   configuration as possible).
 161
 162 <li>avoid resource leaks (memory, shared memory, locks, file
 163   descriptors, X server resources, etc).  Restarting a long-running
 164   server is not always an option.
 165
 166 <li>system services should send messages to syslog.  always log why when
 167   crashing.  always log problems and errors.
 168
 169 <li>reuse configuration.  ktouch have its own x layout setting.
 170   better to fetch the current one from X like xkeycaps.
 171
 172 </ul>
 173
 174     <h2>Solving the upgrade problem using multilevel
 175       configuration</h2>
 176
 177     <ul>
 178
 179       <li>local configuration should be kept during upgrades</li>
 180
 181       <li>easiest to do if the local configration is separate from the
 182         package default
 183
 184       <li>several actors what to have a say in the service
 185          configuration.  allow them to have their own files
 186
 187       <li>Example: read config from /usr/share/foo/config,
 188          /site/share/foo/config, /etc/foo/config, ~/.foo/config,
 189          /etc/foo/config.fixed, /site/share/foo/config.fixed,
 190          /usr/share/foo/config.fixed.
 191
 192       <li>make it possible to provide package, site, host and user
 193         defaults, as well as locking down features on a host, site and
 194         package level.
 195
 196       <li>always well known where the admin made his changes
 197
 198     </ul>
 199
 200
 201
 202     <h2>Thank you very much</h2>
 203
 204     <h3>Questions?</h3>
 205
 206   </body>
 207 </html>