Hi Hoss, Thanks for your reply, Please find answers to your questions below.
*Well, for starters -- have you considered at least looking into using the java based Replicationhandler instead of the rsync scripts?* - There was an attempt to to implement java based replication but it was very slow and so that option was discarded and instead rsync was used. This was done couple of years ago and till Feb of this year, we were using Solr 1.4. I upgraded solr to 4.0 with rsync, however due to time and resource constraint rsync alternative was not evaluated and it can't be done even today - only in next release, we'll be doing solrcloud. My setup looks like below - this was working correctly with Solr 1.4, Solr 4.0 versions. 1) Index Feeder applications feeds indexes to indexer boxes. 2) A cron job that runs every minute on indexer boxes (commiter), commits the indexes (commit) and invokes snapshooter to create snapshot. rsync daemon running on indexer boxes. 3) Another cron job runs on search boxes every minute, which pulls the snapshot (using snappuller), installs it on search boxes (snapinstaller) which also notifies search to open a new searcher (commit) Additionally, there is a cron job that runs every morning at 4 am on indexer boxes which optimises the index (optimize) and cleans the snapshots until a day (snapcleaner). This is as per http://wiki.apache.org/solr/SolrCollectionDistributionScripts *Which config is this, your indexer or your searcher? (i'm assuming it's the searcher since i don't see any postCommit commands to exec snapshooter but i wanted to sanity check that wasn't a simple explanation for your problem)* - Because of this set up, I do not have any post commit setup in solrconfig.xml. - This solrconfig.xml is used for both indexer and searcher boxes. I can see that after my upgrade to Solr 4.2.1, all these scripts behave normally just that I do not see the updates getting refreshed on search boxes unless I restart. * * *What exactly does your "manual commit" command look like? * - This is by using commit script under bin directory (commit -h localhost -p 8983) - I have also tried URL based commit as you had mentioned but no luck *Are you doing this on the indexer box or the searcher boxes? * - I executed manual commit on searcher boxes, the indexer boxes do show the commit and updates correctly. *what is the HTTP response from this comment? what do the logs show when you do this? * - I have attached the logs, please note that I have enabled the openSearcher for testing. Thanks, please let me know if I'm missing something. I remembered people not getting their deletes and the workaround was to add _version_ field in schema, which I had done but no luck. I know it might be unrelated but I am just trying all my options. Thanks again, Sandeep On 5 June 2013 00:41, Chris Hostetter <hossman_luc...@fucit.org> wrote: > > : However, we haven't yet implemented SolrCloud and still relying on > : distribution scripts - rsync, indexpuller mechanism. > > Well, for starters -- have you considered at least looking into using hte > java based Replicationhandler instead of the rsync scripts? > > Script based replication has not been actively maintained since java > replication was added back in Solr 1.4! > > : I see that the indexes are getting created on indexer boxes, snapshots > : being created and then pulled across to search boxes. The snapshots are > : getting installed on search boxes as well. There are no errors in the > : scripts logs and this process works well. > : However, when I check the update in solr console (on search boxes), I do > : not see the updated result. The updates do not appear in search boxes > even > : after manual commit. Only after a *restart* of the search application > : (deployed in tomcat) I can see the updated results. > > What exactly does your "manual commit" command look like? Are you > doing this on the indexer box or the searcher boxes? what is the HTTP > response from this comment? what do the logs show when you do this? > > It's possible that some internal changes in Solr relating to NRT > improvements may have optimized away re-opening on commit if solr doesn't > think the index has changed -- but i doubt it. because I just tried a > simple test using the 4.3.0 example where i manually simulated > snapinstaller replacing hte index files with a newer index and issued > "http://localhost:8983/solr/update?commit=true"; and solr loaded up that > new index and started searching it -- so i suspect the devil is in the > details of your setup. > > you're sure each of the snapshooter, snappuller, snapinstaller scripts are > executing properly? > > : I have done minimal changes for the upgrade in solrconfig.xml and is > pasted > : below. Please can someone take a look and let me know what the issue is. > : The same config was working fine on Solr 4.0 (as well as Solr 1.4.1). > > which config is this, your indexer or your searcher? (i'm assuming it's > the searcher since i don't see any postCommit commands to exec snapshooter > but i wanted to sanity check that wasn't a simple explanation for your > problem) > > -Hoss >
