Metacat Performace: Upgrading hardware setup
Following things have to be done in regards to the upgrading of hardware setup
-> Have to install a new ESA metacat server with Postgres as the db.
-> Upgrade KNB metacat server to ecoinfo2 with Postgres on it.
-> check docs on ecoinfo == docs on ecoinfo2 (This includes the revisions and
the deleted documents. Bug# 2136)
(The above are from the ESA tasklist)
Notes from Matt's email regarding Metacat and ldap setup:
Basically, we want LDAP and metacat to be a failsafe system, which means
eliminating single points of failure. Our proposed solution is to basically
make both NCEAS and LTER house identical copies of all information and services
and act as failover points for each other. The setup is slightly different for
ldap and metacat so I'll talk about both of them separately.
ldap1.ecoinformatics.org resides at NCEAS
master for the NCEAS, UCNRS, and some other trees
slave for LTER trees
slave for KNP tree, PISCO tree, etc
ldap2.ecoinformatics.org resides at LTER
master for LTER
slave for other domains
Matt and James agreed to work with Sid, Duane, and Jing to get this replication
and referral system working.
Uses DNS round-robin to point at both of those servers and redirect clients to
one or the other in a load-balanced and failover way. For this to work we need
to make sure that DNS caching won't delay failover to the other host in event
of a net failure. Colby and Mark will experiment with various DNS
failover/load balanacing techniques and propose a solution to this group.
Basically, same setup as now, except we need to install identical set of skins
and default skin on both the NCEAS and LTER metacats. The NCEAS metacat will
be renamed 'knb1.ecoinformatics.org' and the LTER metacat will
become 'knb2.ecoinformatics.org'. Both of these will be used in a round robin
(as for ldap) based on the address knb.ecoinformatics.org which will be
modified to resolve to both the knb1 and knb2 hosts (and will need to be moved
in a way that doesn't cause downtime).
In normal operation, requests for access will be distributed between both
hosts. In the event of a failure, DNS will redirect hosts to the alternate
address that remains accessible. During a failure, documents with a home
server on the inaccessible host will not be editable (because a lock for the
document will be inaccessible). Otherwise, all other functions will be present.
This will mean that the NCEAS and LTER servers will be treated as a single
integrated system, and upgrades and changes will need to be synchronized
carefull between them. Also, access logs will be split between them because
requests will be regularly routed to both servers.
#3 Updated by Saurabh Garg over 14 years ago
In the Metacat meeting on 27th July, Will pointed out that the configuration of having two metacat servers as failovers of each other could be a problem as we do not know how to handle sessions. A client might start a session with the first server and then the next request from the client might be forwarded to the other server. In this scenario the request sent to second server will fail. Hence we will have to figure out a way of transfering sessions between servers.