<?xml version="1.0" encoding="latin1" ?>
<!DOCTYPE chapter SYSTEM "chapter.dtd">
<chapter>
<header>
<copyright>
<year>2003</year><year>2009</year>
<holder>Ericsson AB. All Rights Reserved.</holder>
</copyright>
<legalnotice>
The contents of this file are subject to the Erlang Public License,
Version 1.1, (the "License"); you may not use this file except in
compliance with the License. You should have received a copy of the
Erlang Public License along with this software. If not, it can be
retrieved online at http://www.erlang.org/.
Software distributed under the License is distributed on an "AS IS"
basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
the License for the specific language governing rights and limitations
under the License.
</legalnotice>
<title>Appup Cookbook</title>
<prepared></prepared>
<docno></docno>
<date></date>
<rev></rev>
<file>appup_cookbook.xml</file>
</header>
<p>This chapter contains examples of <c>.appup</c> files for
typical cases of upgrades/downgrades done in run-time.</p>
<section>
<title>Changing a Functional Module</title>
<p>When a change has been made to a functional module, for example
if a new function has been added or a bug has been corrected,
simple code replacement is sufficient.</p>
<p>Example:</p>
<code type="none">
{"2",
[{"1", [{load_module, m}]}],
[{"1", [{load_module, m}]}]
}.</code>
</section>
<section>
<title>Changing a Residence Module</title>
<p>In a system implemented according to the OTP Design Principles,
all processes, except system processes and special processes,
reside in one of the behaviours <c>supervisor</c>,
<c>gen_server</c>, <c>gen_fsm</c> or <c>gen_event</c>. These
belong to the STDLIB application and upgrading/downgrading
normally requires an emulator restart.</p>
<p>OTP thus provides no support for changing residence modules
except in the case of <seealso marker="#spec">special processes</seealso>.</p>
</section>
<section>
<title>Changing a Callback Module</title>
<p>A callback module is a functional module, and for code
extensions simple code replacement is sufficient.</p>
<p>Example: When adding a function to <c>ch3</c> as described in
the example in <seealso marker="release_handling#appup">Release Handling</seealso>, <c>ch_app.appup</c> looks as follows:</p>
<code type="none">
{"2",
[{"1", [{load_module, ch3}]}],
[{"1", [{load_module, ch3}]}]
}.</code>
<p>OTP also supports changing the internal state of behaviour
processes, see <seealso marker="#int_state">Changing Internal State</seealso> below.</p>
</section>
<section>
<marker id="int_state"></marker>
<title>Changing Internal State</title>
<p>In this case, simple code replacement is not sufficient.
The process must explicitly transform its state using the callback
function <c>code_change</c> before switching to the new version
of the callback module. Thus synchronized code replacement is
used.</p>
<p>Example: Consider the gen_server <c>ch3</c> from the chapter
about the <seealso marker="gen_server_concepts#ex">gen_server behaviour</seealso>. The internal state is a term <c>Chs</c>
representing the available channels. Assume we want add a counter
<c>N</c> which keeps track of the number of <c>alloc</c> requests
so far. This means we need to change the format to
<c>{Chs,N}</c>.</p>
<p>The <c>.appup</c> file could look as follows:</p>
<code type="none">
{"2",
[{"1", [{update, ch3, {advanced, []}}]}],
[{"1", [{update, ch3, {advanced, []}}]}]
}.</code>
<p>The third element of the <c>update</c> instruction is a tuple
<c>{advanced,Extra}</c> which says that the affected processes
should do a state transformation before loading the new version
of the module. This is done by the processes calling the callback
function <c>code_change</c> (see <c>gen_server(3)</c>). The term
<c>Extra</c>, in this case [], is passed as-is to the function:</p>
<marker id="code_change"></marker>
<code type="none">
-module(ch3).
...
-export([code_change/3]).
...
code_change({down, _Vsn}, {Chs, N}, _Extra) ->
{ok, Chs};
code_change(_Vsn, Chs, _Extra) ->
{ok, {Chs, 0}}.</code>
<p>The first argument is <c>{down,Vsn}</c> in case of a downgrade,
or <c>Vsn</c> in case of an upgrade. The term <c>Vsn</c> is
fetched from the 'original' version of the module, i.e.
the version we are upgrading from, or downgrading to.</p>
<p>The version is defined by the module attribute <c>vsn</c>, if
any. There is no such attribute in <c>ch3</c>, so in this case
the version is the checksum (a huge integer) of the BEAM file, an
uninteresting value which is ignored.</p>
<p>(The other callback functions of <c>ch3</c> need to be modified
as well and perhaps a new interface function added, this is not
shown here).</p>
</section>
<section>
<title>Module Dependencies</title>
<p>Assume we extend a module by adding a new interface function, as
in the example in <seealso marker="release_handling#appup">Release Handling</seealso>, where a function <c>available/0</c> is
added to <c>ch3</c>.</p>
<p>If we also add a call to this function, say in the module
<c>m1</c>, a run-time error could occur during release upgrade if
the new version of <c>m1</c> is loaded first and calls
<c>ch3:available/0</c> before the new version of <c>ch3</c> is
loaded.</p>
<p>Thus, <c>ch3</c> must be loaded before <c>m1</c> is, in
the upgrade case, and vice versa in the downgrade case. We say
that <c>m1</c><em>is dependent on</em><c>ch3</c>. In a release
handling instruction, this is expressed by the element
<c>DepMods</c>:</p>
<code type="none">
{load_module, Module, DepMods}
{update, Module, {advanced, Extra}, DepMods}</code>
<p><c>DepMods</c> is a list of modules, on which <c>Module</c> is
dependent.</p>
<p>Example: The module <c>m1</c> in the application <c>myapp</c> is
dependent on <c>ch3</c> when upgrading from "1" to "2", or
downgrading from "2" to "1":</p>
<code type="none">
myapp.appup:
{"2",
[{"1", [{load_module, m1, [ch3]}]}],
[{"1", [{load_module, m1, [ch3]}]}]
}.
ch_app.appup:
{"2",
[{"1", [{load_module, ch3}]}],
[{"1", [{load_module, ch3}]}]
}.</code>
<p>If <c>m1</c> and <c>ch3</c> had belonged to the same application,
the <c>.appup</c> file could have looked like this:</p>
<code type="none">
{"2",
[{"1",
[{load_module, ch3},
{load_module, m1, [ch3]}]}],
[{"1",
[{load_module, ch3},
{load_module, m1, [ch3]}]}]
}.</code>
<p>Note that it is <c>m1</c> that is dependent on <c>ch3</c> also
when downgrading. <c>systools</c> knows the difference between
up- and downgrading and will generate a correct <c>relup</c>,
where <c>ch3</c> is loaded before <c>m1</c> when upgrading but
<c>m1</c> is loaded before <c>ch3</c> when downgrading.</p>
</section>
<section>
<marker id="spec"></marker>
<title>Changing Code For a Special Process</title>
<p>In this case, simple code replacement is not sufficient.
When a new version of a residence module for a special process
is loaded, the process must make a fully qualified call to
its loop function to switch to the new code. Thus synchronized
code replacement must be used.</p>
<note>
<p>The name(s) of the user-defined residence module(s) must be
listed in the <c>Modules</c> part of the child specification
for the special process, in order for the release handler to
find the process.</p>
</note>
<p>Example. Consider the example <c>ch4</c> from the chapter about
<seealso marker="spec_proc#ex">sys and proc_lib</seealso>.
When started by a supervisor, the child specification could look
like this:</p>
<code type="none">
{ch4, {ch4, start_link, []},
permanent, brutal_kill, worker, [ch4]}</code>
<p>If <c>ch4</c> is part of the application <c>sp_app</c> and a new
version of the module should be loaded when upgrading from
version "1" to "2" of this application, <c>sp_app.appup</c> could
look like this:</p>
<code type="none">
{"2",
[{"1", [{update, ch4, {advanced, []}}]}],
[{"1", [{update, ch4, {advanced, []}}]}]
}.</code>
<p>The <c>update</c> instruction must contain the tuple
<c>{advanced,Extra}</c>. The instruction will make the special
process call the callback function <c>system_code_change/4</c>, a
function the user must implement. The term <c>Extra</c>, in this
case [], is passed as-is to <c>system_code_change/4</c>:</p>
<code type="none">
-module(ch4).
...
-export([system_code_change/4]).
...
system_code_change(Chs, _Module, _OldVsn, _Extra) ->
{ok, Chs}.</code>
<p>The first argument is the internal state <c>State</c> passed from
the function <c>sys:handle_system_msg(Request, From, Parent, Module, Deb, State)</c>, called by the special process when
a system message is received. In <c>ch4</c>, the internal state is
the set of available channels <c>Chs</c>.</p>
<p>The second argument is the name of the module (<c>ch4</c>).</p>
<p>The third argument is <c>Vsn</c> or <c>{down,Vsn}</c> as
described for
<seealso marker="#code_change">gen_server:code_change/3</seealso>.</p>
<p>In this case, all arguments but the first are ignored and
the function simply returns the internal state again. This is
enough if the code only has been extended. If we had wanted to
change the internal state (similar to the example in
<seealso marker="#int_state">Changing Internal State</seealso>),
it would have been done in this function and
<c>{ok,Chs2}</c> returned.</p>
</section>
<section>
<marker id="sup"></marker>
<title>Changing a Supervisor</title>
<p>The supervisor behaviour supports changing the internal state,
i.e. changing restart strategy and maximum restart frequency
properties, as well as changing existing child specifications.</p>
<p>Adding and deleting child processes are also possible, but not
handled automatically. Instructions must be given by in
the <c>.appup</c> file.</p>
<section>
<title>Changing Properties</title>
<p>Since the supervisor should change its internal state,
synchronized code replacement is required. However,
a special <c>update</c> instruction must be used.</p>
<p>The new version of the callback module must be loaded first
both in the case of upgrade and downgrade. Then the new return
value of <c>init/1</c> can be checked and the internal state be
changed accordingly.</p>
<p>The following <c>upgrade</c> instruction is used for
supervisors:</p>
<code type="none">
{update, Module, supervisor}</code>
<p>Example: Assume we want to change the restart strategy of
<c>ch_sup</c> from the <seealso marker="sup_princ#ex">Supervisor Behaviour</seealso> chapter from one_for_one to one_for_all.
We change the callback function <c>init/1</c> in
<c>ch_sup.erl</c>:</p>
<code type="none">
-module(ch_sup).
...
init(_Args) ->
{ok, {{one_for_all, 1, 60}, ...}}.</code>
<p>The file <c>ch_app.appup</c>:</p>
<code type="none">
{"2",
[{"1", [{update, ch_sup, supervisor}]}],
[{"1", [{update, ch_sup, supervisor}]}]
}.</code>
</section>
<section>
<title>Changing Child Specifications</title>
<p>The instruction, and thus the <c>.appup</c> file, when
changing an existing child specification, is the same as when
changing properties as described above:</p>
<code type="none">
{"2",
[{"1", [{update, ch_sup, supervisor}]}],
[{"1", [{update, ch_sup, supervisor}]}]
}.</code>
<p>The changes do not affect existing child processes. For
example, changing the start function only specifies how
the child process should be restarted, if needed later on.</p>
<p>Note that the id of the child specification cannot be changed.</p>
<p>Note also that changing the <c>Modules</c> field of the child
specification may affect the release handling process itself,
as this field is used to identify which processes are affected
when doing a synchronized code replacement.</p>
</section>
<marker id="sup_add"></marker>
<section>
<title>Adding And Deleting Child Processes</title>
<p>As stated above, changing child specifications does not affect
existing child processes. New child specifications are
automatically added, but not deleted. Also, child processes are
not automatically started or terminated. Instead, this must be
done explicitly using <c>apply</c> instructions.</p>
<p>Example: Assume we want to add a new child process <c>m1</c> to
<c>ch_sup</c> when upgrading <c>ch_app</c> from "1" to "2".
This means <c>m1</c> should be deleted when downgrading from
"2" to "1":</p>
<code type="none">
{"2",
[{"1",
[{update, ch_sup, supervisor},
{apply, {supervisor, restart_child, [ch_sup, m1]}}
]}],
[{"1",
[{apply, {supervisor, terminate_child, [ch_sup, m1]}},
{apply, {supervisor, delete_child, [ch_sup, m1]}},
{update, ch_sup, supervisor}
]}]
}.</code>
<p>Note that the order of the instructions is important.</p>
<p>Note also that the supervisor must be registered as
<c>ch_sup</c> for the script to work. If the supervisor is not
registered, it cannot be accessed directly from the script.
Instead a help function that finds the pid of the supervisor
and calls <c>supervisor:restart_child</c> etc. must be written,
and it is this function that should be called from the script
using the <c>apply</c> instruction.</p>
<p>If the module <c>m1</c> is introduced in version "2" of
<c>ch_app</c>, it must also be loaded when upgrading and
deleted when downgrading:</p>
<code type="none">
{"2",
[{"1",
[{add_module, m1},
{update, ch_sup, supervisor},
{apply, {supervisor, restart_child, [ch_sup, m1]}}
]}],
[{"1",
[{apply, {supervisor, terminate_child, [ch_sup, m1]}},
{apply, {supervisor, delete_child, [ch_sup, m1]}},
{update, ch_sup, supervisor},
{delete_module, m1}
]}]
}.</code>
<p>Note again that the order of the instructions is important.
When upgrading, <c>m1</c> must be loaded and the supervisor's
child specification changed, before the new child process can
be started. When downgrading, the child process must be
terminated before child specification is changed and the module
is deleted.</p>
</section>
</section>
<section>
<title>Adding or Deleting a Module</title>
<p>Example: A new functional module <c>m</c> is added to
<c>ch_app</c>:</p>
<code type="none">
{"2",
[{"1", [{add_module, m}]}],
[{"1", [{delete_module, m}]}]</code>
</section>
<section>
<title>Starting or Terminating a Process</title>
<p>In a system structured according to the OTP design principles,
any process would be a child process belonging to a supervisor,
see <seealso marker="#sup_add">Adding and Deleting Child Processes</seealso> above.</p>
</section>
<section>
<title>Adding or Removing an Application</title>
<p>When adding or removing an application, no <c>.appup</c> file
is needed. When generating <c>relup</c>, the <c>.rel</c> files
are compared and <c>add_application</c> and
<c>remove_application</c> instructions are added automatically.</p>
</section>
<section>
<title>Restarting an Application</title>
<p>Restarting an application is useful when a change is too
complicated to be made without restarting the processes, for
example if the supervisor hierarchy has been restructured.</p>
<p>Example: When adding a new child <c>m1</c> to <c>ch_sup</c>, as
in the <seealso marker="#sup_add">example above</seealso>, an
alternative to updating the supervisor is to restart the entire
application:</p>
<code type="none">
{"2",
[{"1", [{restart_application, ch_app}]}],
[{"1", [{restart_application, ch_app}]}]
}.</code>
</section>
<section>
<marker id="app_spec"></marker>
<title>Changing an Application Specification</title>
<p>When installing a release, the application specifications are
automatically updated before evaluating the <c>relup</c> script.
Hence, no instructions are needed in the <c>.appup</c> file:</p>
<pre>
{"2",
[{"1", []}],
[{"1", []}]
}.</pre>
</section>
<section>
<title>Changing Application Configuration</title>
<p>Changing an application configuration by updating the <c>env</c>
key in the <c>.app</c> file is an instance of changing an
application specification, <seealso marker="#app_spec">see above</seealso>.</p>
<p>Alternatively, application configuration parameters can be
added or updated in <c>sys.config</c>.</p>
</section>
<section>
<title>Changing Included Applications</title>
<p>The release handling instructions for adding, removing and
restarting applications apply to primary applications only.
There are no corresponding instructions for included
applications. However, since an included application is really a
supervision tree with a topmost supervisor, started as a child
process to a supervisor in the including application, a
<c>relup</c> file can be manually created.</p>
<p>Example: Assume we have a release containing an application
<c>prim_app</c> which have a supervisor <c>prim_sup</c> in its
supervision tree.</p>
<p>In a new version of the release, our example application
<c>ch_app</c> should be included in <c>prim_app</c>. That is,
its topmost supervisor <c>ch_sup</c> should be started as a child
process to <c>prim_sup</c>.</p>
<p>1) Edit the code for <c>prim_sup</c>:</p>
<code type="none">
init(...) ->
{ok, {...supervisor flags...,
[...,
{ch_sup, {ch_sup,start_link,[]},
permanent,infinity,supervisor,[ch_sup]},
...]}}.</code>
<p>2) Edit the <c>.app</c> file for <c>prim_app</c>:</p>
<code type="none">
{application, prim_app,
[...,
{vsn, "2"},
...,
{included_applications, [ch_app]},
...
]}.</code>
<p>3) Create a new <c>.rel</c> file, including <c>ch_app</c>:</p>
<code type="none">
{release,
...,
[...,
{prim_app, "2"},
{ch_app, "1"}]}.</code>
<section>
<title>Application Restart</title>
<p>4a) One way to start the included application is to restart
the entire <c>prim_app</c> application. Normally, we would then
use the <c>restart_application</c> instruction in
the <c>.appup</c> file for <c>prim_app</c>.</p>
<p>However, if we did this and then generated a <c>relup</c> file,
not only would it contain instructions for restarting (i.e.
removing and adding) <c>prim_app</c>, it would also contain
instructions for starting <c>ch_app</c> (and stopping it, in
the case of downgrade). This is due to the fact that
<c>ch_app</c> is included in the new <c>.rel</c> file, but not
in the old one.</p>
<p>Instead, a correct <c>relup</c> file can be created manually,
either from scratch or by editing the generated version.
The instructions for starting/stopping <c>ch_app</c> are
replaced by instructions for loading/unloading the application:</p>
<code type="none">
{"B",
[{"A",
[],
[{load_object_code,{ch_app,"1",[ch_sup,ch3]}},
{load_object_code,{prim_app,"2",[prim_app,prim_sup]}},
point_of_no_return,
{apply,{application,stop,[prim_app]}},
{remove,{prim_app,brutal_purge,brutal_purge}},
{remove,{prim_sup,brutal_purge,brutal_purge}},
{purge,[prim_app,prim_sup]},
{load,{prim_app,brutal_purge,brutal_purge}},
{load,{prim_sup,brutal_purge,brutal_purge}},
{load,{ch_sup,brutal_purge,brutal_purge}},
{load,{ch3,brutal_purge,brutal_purge}},
{apply,{application,load,[ch_app]}},
{apply,{application,start,[prim_app,permanent]}}]}],
[{"A",
[],
[{load_object_code,{prim_app,"1",[prim_app,prim_sup]}},
point_of_no_return,
{apply,{application,stop,[prim_app]}},
{apply,{application,unload,[ch_app]}},
{remove,{ch_sup,brutal_purge,brutal_purge}},
{remove,{ch3,brutal_purge,brutal_purge}},
{purge,[ch_sup,ch3]},
{remove,{prim_app,brutal_purge,brutal_purge}},
{remove,{prim_sup,brutal_purge,brutal_purge}},
{purge,[prim_app,prim_sup]},
{load,{prim_app,brutal_purge,brutal_purge}},
{load,{prim_sup,brutal_purge,brutal_purge}},
{apply,{application,start,[prim_app,permanent]}}]}]
}.</code>
</section>
<section>
<title>Supervisor Change</title>
<p>4b) Another way to start the included application (or stop it
in the case of downgrade) is by combining instructions for
adding and removing child processes to/from <c>prim_sup</c> with
instructions for loading/unloading all <c>ch_app</c> code and
its application specification.</p>
<p>Again, the <c>relup</c> file is created manually. Either from
scratch or by editing a generated version. Load all code for
<c>ch_app</c> first, and also load the application
specification, before <c>prim_sup</c> is updated. When
downgrading, <c>prim_sup</c> should be updated first, before
the code for <c>ch_app</c> and its application specification
are unloaded.</p>
<code type="none">
{"B",
[{"A",
[],
[{load_object_code,{ch_app,"1",[ch_sup,ch3]}},
{load_object_code,{prim_app,"2",[prim_sup]}},
point_of_no_return,
{load,{ch_sup,brutal_purge,brutal_purge}},
{load,{ch3,brutal_purge,brutal_purge}},
{apply,{application,load,[ch_app]}},
{suspend,[prim_sup]},
{load,{prim_sup,brutal_purge,brutal_purge}},
{code_change,up,[{prim_sup,[]}]},
{resume,[prim_sup]},
{apply,{supervisor,restart_child,[prim_sup,ch_sup]}}]}],
[{"A",
[],
[{load_object_code,{prim_app,"1",[prim_sup]}},
point_of_no_return,
{apply,{supervisor,terminate_child,[prim_sup,ch_sup]}},
{apply,{supervisor,delete_child,[prim_sup,ch_sup]}},
{suspend,[prim_sup]},
{load,{prim_sup,brutal_purge,brutal_purge}},
{code_change,down,[{prim_sup,[]}]},
{resume,[prim_sup]},
{remove,{ch_sup,brutal_purge,brutal_purge}},
{remove,{ch3,brutal_purge,brutal_purge}},
{purge,[ch_sup,ch3]},
{apply,{application,unload,[ch_app]}}]}]
}.</code>
</section>
</section>
<section>
<title>Changing Non-Erlang Code</title>
<p>Changing code for a program written in another programming
language than Erlang, for example a port program, is very
application dependent and OTP provides no special support for it.</p>
<p>Example, changing code for a port program: Assume that
the Erlang process controlling the port is a gen_server
<c>portc</c> and that the port is opened in the callback function
<c>init/1</c>:</p>
<code type="none">
init(...) ->
...,
PortPrg = filename:join(code:priv_dir(App), "portc"),
Port = open_port({spawn,PortPrg}, [...]),
...,
{ok, #state{port=Port, ...}}.</code>
<p>If the port program should be updated, we can extend the code for
the gen_server with a <c>code_change</c> function which closes
the old port and opens a new port. (If necessary, the gen_server
may first request data that needs to be saved from the port
program and pass this data to the new port):</p>
<code type="none">
code_change(_OldVsn, State, port) ->
State#state.port ! close,
receive
{Port,close} ->
true
end,
PortPrg = filename:join(code:priv_dir(App), "portc"),
Port = open_port({spawn,PortPrg}, [...]),
{ok, #state{port=Port, ...}}.</code>
<p>Update the application version number in the <c>.app</c> file
and write an <c>.appup</c> file:</p>
<code type="none">
["2",
[{"1", [{update, portc, {advanced,port}}]}],
[{"1", [{update, portc, {advanced,port}}]}]
].</code>
<p>Make sure the <c>priv</c> directory where the C program is
located is included in the new release package:</p>
<pre>
1> <input>systools:make_tar("my_release", [{dirs,[priv]}]).</input>
...</pre>
</section>
<section>
<title>Emulator Restart</title>
<p>If the emulator can or should be restarted, the very simple
<c>.relup</c> file can be created manually:</p>
<code type="none">
{"B",
[{"A",
[],
[restart_new_emulator]}],
[{"A",
[],
[restart_new_emulator]}]
}.</code>
<p>This way, the release handler framework with automatic packing
and unpacking of release packages, automatic path updates etc. can
be used without having to specify <c>.appup</c> files.</p>
<p>If some transformation of persistent data, for example database
contents, needs to be done before installing the new release
version, instructions for this can be added to the <c>.relup</c>
file as well.</p>
</section>
</chapter>
