Hi all, As a step in SOLR-8282, I'm trying to get all access to the data directory done by Solr to be mediated through the DirectoryFactory implementation. Part of this is the creation of the UpdateLog, and I'm a bit confused by some of the logic in there currently.
The UpdateLog is created by the UpdateHandler, which has some logic in there to determine whether or not to use a standard log or an HDFSUpdateLog. In particular, around line 117, we check to see if the update log directory begins with "hdfs:/", and if it does we then do a further check to see if the directory factory is an HDFSDirectoryFactory or not. This seems to imply that Solr currently supports storing the update log in HDFS even if the actual indexes are on a normal file system. Which seems odd, at the very least. All our docs say to use HDFSDirectoryFactory if you want to store anything in HDFS, and there's nothing anywhere about storing the update logs separately from the indexes. Is this a relic of past behaviour, or is it something that a) should be preserved by the refactoring I'm doing, and b) documented and tested? Alan Woodward www.flax.co.uk
