Activity log for bug #931604

Date Who What changed Old value New value Message
2012-02-13 18:49:59 Armando Migliaccio bug added bug
2012-02-13 18:50:07 Armando Migliaccio nova: assignee Armando Migliaccio (armando-migliaccio)
2012-02-13 18:50:09 Armando Migliaccio nova: status New Confirmed
2012-02-13 18:53:52 OpenStack Infra nova: status Confirmed In Progress
2012-02-13 18:57:12 Armando Migliaccio description There is nothing particularly wrong with how RRD records are retrieved in xenapi today. gi However there are a couple of potential issues, as described below: -RRD XMLs are always obtained using http over the XS management interface address. Albeit this is fine in most cases there are problems when: i) nova-compute domU is talking to xapi using the internal management network (e.g. talking to xapi via 169.254.0.1) and traffic cannot be routed between domU and dom0; ii) in resource pools, the host address from the xenapi session handle is always the master, which could get overloaded if there are too many requests. -get_rrd and get_rrd_updates could politely log the IOError exceptions. I propose to use FLAGS.xenapi_connection_url instead. By using this flag we have the following benefits: - we transparently use the appropriate scheme and server (regardless of whether the host is a pool or not). - it is quicker (we don't need to call xapi to get the address) There is nothing particularly wrong with how RRD records are retrieved in xenapi today. However there are a couple of potential issues, as described below: -RRD XMLs are always obtained using http over the XS management interface address. Albeit this is fine in most cases there are problems when: i) nova-compute domU is talking to xapi using the internal management network (e.g. talking to xapi via 169.254.0.1) and traffic cannot be routed between domU and dom0; ii) in resource pools, the host address from the xenapi session handle is always the master, which could get overloaded if there are too many requests. - get_rrd and get_rrd_updates could politely log the IOError exceptions. I propose to use FLAGS.xenapi_connection_url instead. By using this flag we have the following benefits: - we transparently use the appropriate scheme and server (regardless of whether the host is a pool or not). - it is quicker (we don't need to call xapi to get the address)
2012-02-13 18:57:48 Armando Migliaccio description There is nothing particularly wrong with how RRD records are retrieved in xenapi today. However there are a couple of potential issues, as described below: -RRD XMLs are always obtained using http over the XS management interface address. Albeit this is fine in most cases there are problems when: i) nova-compute domU is talking to xapi using the internal management network (e.g. talking to xapi via 169.254.0.1) and traffic cannot be routed between domU and dom0; ii) in resource pools, the host address from the xenapi session handle is always the master, which could get overloaded if there are too many requests. - get_rrd and get_rrd_updates could politely log the IOError exceptions. I propose to use FLAGS.xenapi_connection_url instead. By using this flag we have the following benefits: - we transparently use the appropriate scheme and server (regardless of whether the host is a pool or not). - it is quicker (we don't need to call xapi to get the address) There is nothing particularly wrong with how RRD records are retrieved in xenapi today. However there are a couple of potential issues, as described below: - RRD XMLs are always obtained using http over the XS management interface address. Albeit this is fine in most cases there are problems when: i) nova-compute domU is talking to xapi using the internal management network (e.g. talking to xapi via 169.254.0.1) and traffic cannot be routed between domU and dom0; ii) in resource pools, the host address from the xenapi session handle is always the master, which could get overloaded if there are too many requests. - get_rrd and get_rrd_updates could politely log the IOError exceptions. I propose to use FLAGS.xenapi_connection_url instead. By using this flag we have the following benefits: - we transparently use the appropriate scheme and server (regardless of whether the host is in a pool or not). - it is quicker (we don't need to call xapi to get the address)
2012-02-17 00:16:33 OpenStack Infra nova: status In Progress Fix Committed
2012-02-29 10:27:20 Thierry Carrez nova: status Fix Committed Fix Released
2012-02-29 10:27:20 Thierry Carrez nova: milestone essex-4
2012-04-05 10:14:11 Thierry Carrez nova: milestone essex-4 2012.1