mail to sakai-dev:
This afternoon we suffred a period of high utalization on our database back end for a period. I'm still investigating to see what caused it but it seelms that at least part of this was a high number of long running query for realm grants:
SELECT SAKAI_REALM_ROLE.ROLE_NAME, SAKAI_REALM_RL_GR.USER_ID, SAKAI_REALM_RL_GR.ACTIVE, SAKAI_REALM_RL_GR.PROVIDED FROM SAKAI_REALM_RL_GR INNER JOIN SAKAI_REALM ON SAKAI_REALM.REALM_KEY = SAKAI_REALM_RL_GR.REALM_KEY AND SAKAI_REALM.REALM_ID = '/site/96dbcba6-01c2-4059-007a-40e484873bb9' INNER JOIN SAKAI_REALM_ROLE ON SAKAI_REALM_ROLE.ROLE_KEY = SAKAI_REALM_RL_GR.ROLE_KEY;
adding a quick cache and visiting roster and site info in a site resulted in 1 miss and 8 hits on this cache. Now while it is easy to add a cache at this point the AuthzGroup code seems to assume that this information is not cached, and making sure we invalidate the cache in a cluster safe way is slightly more complex....