Hoss:Your are right. It has a version byte written first. This can be used for any changes that come later..So , when we introduce any change to the format we can rely on that. If/When we upgrade the format we must ensure that it is backward compatible .
The format can be used by SolrJ clients as well and they can get all the perf benefits. I shall document it in the wiki. --Noble On Mon, Jun 2, 2008 at 1:44 AM, Chris Hostetter <[EMAIL PROTECTED]> wrote: > > : This comment for the benefit of who is using distributed search: > > : The protocol of communication has been xml for distributed search. For > : a good part of 1.3. > : It is now changed to a custom binary format (SOLR-486 ). So each shard > : participating in a distributed search must be using the same protocol. > : Safest thing is to install each shard with same binaries. > > presumably the binary format includes a header with version info, so > future enhancements to the format can be backwards compatible, correct? > (the same way new versions of Lucene can open indexes created by older > versions) > > it would really suck if the only way to upgrade a distributed search > cluster in the future was to take the whole cluster offline an upgrade > them all at once ... we want to make sure that in the future people > upgrading from 1.3 to 1.X can bring up a new "aggregator" running > 1.X, but pointing at all the same shards (running 1.3) and then upgrade > the individual shards one at a time. > > (either that, or the aggregator needs to specify the max format "version" > it can handle when querying the shards, and then the shards could be > upgraded first) > > > -Hoss > > -- --Noble Paul
