Thanks a lot.
On Sat, Jun 15, 2019 at 11:33 AM Shawn Heisey via dovecot < dovecot@dovecot.org> wrote:
On 6/14/2019 2:11 AM, luckydog xf via dovecot wrote:
And based on https://dovecot.org/pipermail/dovecot/2019-April/115575.html I'm going to use an VIP to host 2 mail servers. Currently, it works in fail over and fail back test except solr index, so how to resolve this?
Solr has high availability options. Probably the easiest is SolrCloud, because the legacy master-slave replication option basically has the master as a single point of failure. Switching a slave to master is a very manual job. SolrCloud is a true cluster -- no masters and no slaves.
Normally I would suggest a load balancer in front of Solr, but with SolrCloud and this usage, you could have a VIP switch between two hosts. You will need at least three servers for a redundant SolrCloud install, because ZooKeeper absolutely requires three servers for redundancy. SolrCloud is Solr+ZooKeeper. Only two of the servers would need to run Solr, though. The third server would only need to run ZooKeeper, and normally the system requirements for that are very low.
Is it possible to put dovecot index on shared storage like NFS?
Solr does not do well on network file systems. They usually don't offer the file locking capability that Lucene wants. Solr is written using the Lucene API.
And you can't have two running Solr servers using the same index directory even if you disable file locking so that NFS works. Each server needs its own copy of the index.
How to rebuild if I don't put index data on shared storage in case of fail over?
Building and rebuilding is accomplished by dovecot. Here's a wiki page for rebuilding Solr indexes. Admittedly it falls into the "not all that helpful" category:
https://wiki.apache.org/solr/HowToReindex
Thanks, Shawn