attempting to retrieve an invalid MARC record via Z39.50 can result in unexpected memory consumption

[Galen Charlton]

Z39.50 targets in the wild have been known to return MARC records that are invalid by virtue of their being longer than 99,999 octets. In at least one case, attempting to retrieve such a record has resulted in an open-ils.search drone ballooning to 3G+ of memory usage during a stage where XML::LibXML is trying to parse a MARCXML version of the record.

An example can be found by searching for

@attr 1=4 @attr 4=1 @attr 5=1 "three can play that game"

at the target z3950.fcla.edu/RF. As of the date of filing this bug, the second hit in the result set triggers the memory ballooning.

Evergreen master

[Galen Charlton (gmc)]

A patch is available in the user/gmcharlt/lp1358916_dont_fetch_overlarge_records_via_z3950 branch of the working/Evergreen repository:

http://git.evergreen-ils.org/?p=working/Evergreen.git;a=shortlog;h=refs/heads/user/gmcharlt/lp1358916_dont_fetch_overlarge_records_via_z3950

[Galen Charlton (gmc)]

The patch is now in production at one consortium and appears to have the desired effect. An open question is whether to adjust the hit count to exclud rejected records.

[Ben Shum (bshum)]

Pushed to master and backported to rel_2_6 and rel_2_5 as a bug fix. Thanks Galen! Silly MARC....