Flume should include hadoop's class path jars.

The compression codec (FLUME-132) was finding codec's that weren't necessarily in flume's class path.

In this particular case the LZO algorithm jar was in the hadoop classpath but not in flume's. When flume asked for codecs, it failed to find the lzo codec and the collector then failed.

A work around a was to copy the lzo jar into flume's lib dir. Ideally, the "missing" codec would be a recoverable warning, or any hadoop jars would be added to flume's jar path. The latter may cause on jar conflict pain however.