As of 31.05.12 the GOCDB service is provided through two portals: a read only instance at goc.egi.eu and a read/write instance at gocdb4.esc.rl.ac.uk. We would like to consolidate these and provide the service (including PI) through one read/write portal at goc.egi.eu.
Two portals were provided to support the deployment of multiple "regional" read/write portals containing different data. These "regional" portals would publish their data to the "central" read-only portal. A central read/write portal was provided for NGIs who didn't want to host their own regional GOCDB instance. However as a regional synchronised GOCDB doesn't yet exist this separation confuses users.
If a regional synchronising GOCDB is developed in future we can revert back to the old model.
- Bring up the planned changes at JRA-1 (Done 24.05.12)
- Create an RT ticket from the proposal and send it round OTAG, JRA-1 and SA-1.
- Agree on a date
- Widely disseminate the change beforehand, giving at least one month's notice.
- No more gocdb4.esc.rl.ac.uk
- Switch ops-monitor to check goc.egi.eu instead of gocdb4.esc.rl.ac.uk.
- Make the change
Switch off gocdb4.esc.rl.ac.uk and switch goc.egi.eu to read/write mode.
Preparing for the Switch
- Prepare a GOCDB update with the following features:
- Remove the "Portal Instance" user interface element
- Remove landing page
- Change the front page to refer to only one "GOCDB"
- Move the GOCDB monitor to goc.egi.eu (/testportal/GOCDB_monitor/ops_monitor_check.php)
- Point ops-monitor's Nagios check "ops_monitor_check.php" at goc.egi.eu rather than gocdb4.esc.rl.ac.uk
- Point SCT's internal Nagios at goc.egi.eu rather than gocdb4.esc.rl.ac.uk
On the Day of the Switch
- Switch goc.egi.eu to read/write mode
- Disable ops-monitor Nagios checks entirely for gocdb4.esc.rl.ac.uk
- Disable SCT's internal Nagios entirely for gocdb4.esc.rl.ac.uk.
- Power down gocdb4.esc.rl.ac.uk, save logs
- Remove gocdb4.esc.rl.ac.uk from DNS
(Note: SCT is the group that hosts GOCDB within STFC).
In this plan, the load on the read/write portal and the read-only portal would be placed on the same machine. This could potentially affect the service.
The graphs below show one week's worth of CPU load for both machines. As this load is so low we believe this won't be an issue.
Note: the regular orange bars show a high amount of wait on IO. This is caused by our daily backup procedure.