I specified -Djava.io.tmpdir=/my/big/partition/path to PIG_OPTS
and I can see that this is indeed set on the JVM args,
but when I ran
pig -x local my_pig_script
it still dumped temp files into /tmp, instead of the custom dir I specified.
I tried adding
SET mapred.child.java.opts '-Djava.io.tmpdir=/my/big/partition/path'
to the pig script too (this is 0.9.2 so the SET works), but it still did
not use the custom dir.
I suspect that for the LocalRunner, hadoop has a bug so it uses the
hard-coded /tmp dir instead of querying the JVM arg