http://www.perlmonks.org?node_id=654153


in reply to Re^2: Memcache 'get' returns undef running under mod_perl (debug)
in thread Memcache 'get' returns undef running under mod_perl

To continue with the simple and obvious suggestions (that you may have already tried), add "-vv" to your memcached command line. (I was also suspicious of your "-u root" as hsinclai highlighted but it'd also be interesting to learn of the mechanics of the failure.)

- tye        

Replies are listed 'Best First'.
Re^4: Memcache 'get' returns undef running under mod_perl (debug)
by graq (Curate) on Dec 03, 2007 at 10:05 UTC

    The problem lay in the FilterHandler's invocation of other classes.

    The code looked a little like this

    $html =~ s/<\?KEYWORD\s+([^\?^>]+)\??>/$self->handle($1)/gme;

    Changing this to

    $html =~ s/<\?KEYWORD\s+([^\?^>]+)\??>/$self->handle($1)/ge; # Removed multiline option 'm'

    fixed the memcache problem.

    It seems to alter memcache's behaviour in GetParser when checking $self->[BUF]

    if ($self->[BUF] =~ /^END\r\n/) { $self->[ON_ITEM] = undef; return 1; # we're done successfully, return 1 to finish }

    Looks like m fighting with S*

    -=( Graq )=-

      To be honest, that makes no sense. /m only impacts the meaning of ^ and the only '^'s in the regex you show are inside a character class and so aren't impacted by /m.

      If this change fixed the mod_perl client, then it does remind me of mod_perl bugs I've seen before. Usually these bugs are heisenbugs, however, coming and going depending on how long a given Apache process has been running, though I guess I have seen some persistent ones (and rarely ones that persisted until the master Apache process was restarted).

      In any case, bug(s) I've seen cause rather weird regex behavior. For example, /\d\d/g will sometimes fail on a string full of pairs of digits (and you'll see a node temporarily tagged as being written "never"). Or I've had several complex regexes that mod_perl refused to parse even though they were correct and worked outside of mod_perl. Changing to a different delimiter was the most common solution.

      Although I don't understand your explanation of the workings of this bug, it certainly looks like a Perl bug if dropping /m does what you describe (note that Perl bugs are more likely to be visible in mod_perl because the Perl interpretter runs for so long). It sounds like perhaps the /m on the one regex is, in mod_perl, making the /^END\r\n/ behave as if it had a /m on it.

      - tye        

        Interesting. Testing it a little further, it appears that it is the combination /me causing the problems. Removing just the 'm' to give

        $html =~ s/<\?KEYWORD\s+([^\?^>]+)\??>/$self->handle($1)/ge;

        does make it 'work'.

        However, I need the match to be multiline. Re-introducing the /m option in a loop also makes things work:

        my $MAX = 100; while( $html =~ m/<\?KEYWORD\s+([^\?^>]+)\??>/m ) { my $ssi_html = $self->handle($1); $html =~ s/<\?KEYWORD\s+([^\?^>]+)\??>/$ssi_html/m; last if $MAX++ > 100; }

        -=( Graq )=-

      Sir, I don't know where is the FilterHandler file is. Could please tell me?