On our large website, I generate many indexes for the swish-e search
engine to use to obtain results. While generating one of the many
indexes, I am now getting the following error...
err: Buffer too short in coalesce_word_locations. Increase
COALESCE_BUFFER_MAX_SIZE in config.h and rebuild.
I'm on a Linux OS Kernel 2.6.9-22 with Swish-e 2.1-dev-24.
The error indicates either a binary file is being parsed or simply "too
many words". The directory that this one index is using has over 19,000
documents. If I remove several subdirectories (over 5,000 documents),
the index is generated (12 MB), but not as large as the last know
working index (14 MB) before I noticed this problem.
There is a binary file that is being parsed. I have had no problems
with this binary file being there in the past years. My guess is that I
have "too many words". Users are constanctly adding and modifying
files. A much smaller directory that only has a bit over 5,000
documents, generates an index of over 32 MB. I have no problem with
this directory or index.
I have gone into the ./src/config.h file and bumped up the value of 256
MB to 1024 for COALESCE_BUFFER_MAX_SIZE, recompiled and I still get the
same error. The comments state that "COALESCE_BUFFER_MAX_SIZE should
not be altered.
Besides breaking up the large directory into 2 and having 2 indexes, is
there another option? Breaking up the directory would cause a lot of
work to recode the html and other files.
Thanks in advance,
Steve
Received on Mon Nov 14 16:41:31 2005