nix-super/doc/manual/troubleshooting.xml
2007-11-29 14:15:33 +00:00

202 lines
6.3 KiB
XML
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

<appendix xmlns="http://docbook.org/ns/docbook"
xmlns:xlink="http://www.w3.org/1999/xlink">
<title>Troubleshooting</title>
<para>This section provides solutions for some common problems. See
the <link xlink:href="https://bugs.cs.uu.nl/browse/NIX">Nix
bug tracker</link> for a list of currently known issues.</para>
<section><title>Berkeley DB: <quote>Cannot allocate memory</quote></title>
<para>Symptom: Nix operations (in particular the
<command>nix-store</command> operations <option>--gc</option>,
<option>--verify</option>, and <option>--clear-substitutes</option>
the latter being called by <command>nix-channel --update</command>)
failing:
<screen>
$ nix-store --verify
error: Db::del: Cannot allocate memory</screen>
</para>
<para>Possible solution: make sure that no Nix processes are running,
then do:
<screen>
$ cd /nix/var/nix/db
$ rm __db.00*</screen>
</para>
</section>
<section><title>Berkeley DB gives weird error messages</title>
<para>Symptom: you get error messages such as
<screen>
Berkeley DB message: Finding last valid log LSN: file: 1 offset 28
Berkeley DB error: file validpaths (meta pgno = 0) has LSN [483][34721].
Berkeley DB error: end of log is [1][28]
Berkeley DB error: /nix/var/nix/db/validpaths: unexpected file type or format</screen>
or other weird Berkeley DB errors, and they dont go away (i.e.,
automatic recovery doesnt work). This may be the case after a system
crash.</para>
<para>Solution: first try to run <command>db_recover</command> and
then <link linkend='refsec-nix-store-verify'><command>nix-store
--verify</command></link>:
<screen>
$ db_recover -h /nix/var/nix/db
$ nix-store --verify</screen>
(Make sure that you have the right version of
<command>db_recover</command>, namely, Berkeley DB 4.4 for Nix 0.10,
and 4.5 for Nix 0.11.)</para>
<para>If that doesnt work, its time to bring out the big guns:
<screen>
$ cd /nix/var/nix
$ cp -pr db db-backup <lineannotation>(making a backup just in case)</lineannotation>
$ cd db
$ rm __db.* log* <lineannotation>(removing the Berkeley DB environment)</lineannotation>
$ mkdir tmp
$ for i in *; do db_dump $i | (cd tmp &amp;&amp; db_load $i); done
<lineannotation>(ignore error messages about non-database files like “reserved”)</lineannotation>
$ mv tmp/* .
$ nix-store --verify</screen>
</para>
</section>
<section><title>Berkeley DB out of locks</title>
<para>It is possible, especially in <command>nix-store
--verify</command> or when running the garbage collector, to run out
of Berkeley DB locks, like this:
<screen>
$ nix-store --verify
checking path existence
checking path realisability
checking the derivers table
checking the references table
Berkeley DB error: Lock table is out of available object entries
error: Db::get: Cannot allocate memory</screen>
</para>
<para>A workaround is to increase the number of locks that Berkeley DB
allocates. (The real solution would be for Nix to not use so many
locks.) This can be done by putting the following in the file
<filename>/nix/var/nix/db/<link
xlink:href="http://www.oracle.com/technology/documentation/berkeley-db/db/ref/env/db_config.html">DB_CONFIG</link></filename>:
<programlisting>
set_lk_max_locks 100000
set_lk_max_lockers 100000
set_lk_max_objects 100000
</programlisting>
(Increase these numbers if necessary.) Then make sure that there are
no running Nix processes and delete the Berkeley DB environment:
<screen>
$ rm /nix/var/nix/db/__db.*</screen>
The Berkeley DB environment is automatically recreated with the new
limits when you run any Nix command.</para>
</section>
<section><title>Collisions in <command>nix-env</command></title>
<para>Symptom: when installing or upgrading, you get an error message such as
<screen>
$ nix-env -i docbook-xml
...
adding /nix/store/s5hyxgm62gk2...-docbook-xml-4.2
collission between `/nix/store/s5hyxgm62gk2...-docbook-xml-4.2/xml/dtd/docbook/calstblx.dtd'
and `/nix/store/06h377hr4b33...-docbook-xml-4.3/xml/dtd/docbook/calstblx.dtd'
at /nix/store/...-builder.pl line 62.</screen>
</para>
<para>The cause is that two installed packages in the user environment
have overlapping filenames (e.g.,
<filename>xml/dtd/docbook/calstblx.dtd</filename>. This usually
happens when you accidentally try to install two versions of the same
package. For instance, in the example above, the Nix Packages
collection contains two versions of <literal>docbook-xml</literal>, so
<command>nix-env -i</command> will try to install both. The default
user environment builder has no way to way to resolve such conflicts,
so it just gives up.</para>
<para>Solution: remove one of the offending packages from the user
environment (if already installed) using <command>nix-env
-e</command>, or specify exactly which version should be installed
(e.g., <literal>nix-env -i docbook-xml-4.2</literal>).</para>
<para>Alternatively, you can modify the user environment builder
script (in
<filename><replaceable>prefix</replaceable>/share/nix/corepkgs/buildenv/builder.pl</filename>)
to implement some conflict resolution policy. E.g., the script could
be modified to rename conflicting file names, or to pick one over the
other.</para>
</section>
<section><title><quote>Too many links</quote> error in the Nix
store</title>
<para>Symptom: when building something, you get an error message such as
<screen>
...
<literal>mkdir: cannot create directory `/nix/store/<replaceable>name</replaceable>': Too many links</literal></screen>
</para>
<para>This is usually because you have more than 32,000 subdirectories
in <filename>/nix/store</filename>, as can be seen using <command>ls
-l</command>:
<screen>
$ ls -l /nix/store
drwxrwxrwt 32000 nix nix 4620288 Sep 8 15:08 store</screen>
The <literal>ext2</literal> file system is limited to a inode link
count of 32,000 (each subdirectory increasing the count by one).
Furthermore, the <literal>st_nlink</literal> field of the
<function>stat</function> system call is a 16-bit value.</para>
<para>This only happens on very large Nix installations (such as build
machines).</para>
<para>Quick solution: run the garbage collector.</para>
<para>Real solution: put the Nix store on a file system that supports
more than 32,000 subdirectories per directory, such as ReiserFS.
(This doesnt solve the <literal>st_nlink</literal> limit, but
ReiserFS lies to the kernel by reporting a link count of 1 if it
exceeds the limit.)</para>
</section>
</appendix>