Difference between revisions of "Pci resource management"
| Line 11: | Line 11: | ||
<sub> | <sub> | ||
| − | problem 1: | + | *problem 1: |
| − | 'claimed' status must be used to keep old device on source machine. | + | **'claimed' status must be used to keep old device on source machine. |
we use node_id, not hostname, so use claimed status to keep out other machine's device out. | we use node_id, not hostname, so use claimed status to keep out other machine's device out. | ||
| − | solution A: get compute node from instance's hostname --> get compute_node_id | + | **solution A: get compute node from instance's hostname --> get compute_node_id |
risk: always reading DB can not cache. | risk: always reading DB can not cache. | ||
| − | 方法2: keep claimed status | + | **方法2: keep claimed status |
| − | problem 2: | + | *problem 2: |
| − | + | ** multiple objects for same db instance of pci devices | |
| − | + | ** must operating the local dev(otherwise miss the save process) | |
</sub> | </sub> | ||
Revision as of 03:23, 20 January 2015
a) Same PCI had different Objects, one in the stats and another in the instance
b) when RESIZE secondary host get other instances Objects
Resize:
-> new host claim -> old host stop (instance going to new HOST)-> new host starting new instance
Confirm:
new machine call source machines’s drop_resize_claim ( a new resource tracker been setting up)
Aborting:
new: stop new instance, drop claim -- > old: reverting old status and stating machine
- problem 1:
- 'claimed' status must be used to keep old device on source machine.
we use node_id, not hostname, so use claimed status to keep out other machine's device out.
- solution A: get compute node from instance's hostname --> get compute_node_id
risk: always reading DB can not cache.
- 方法2: keep claimed status
- problem 2:
** multiple objects for same db instance of pci devices ** must operating the local dev(otherwise miss the save process)
