What was the value you had for host.reservation.release.period before changing? Did you restart the management server after changing the value? I can only see this option to create an issue. Check in the logs when you tried to create a new VM the capacity values that the mgmt server holds, or the value stored in the DB in "op_host_capacity".
On Wed, Sep 27, 2017 at 11:34 PM, Jochim, Ingo <[email protected]> wrote: > It's not about deleted VM's. It's about VM's in shutdown mode. > We do have customers which keep some demo or testing machines in off mode > over a long period. > Off machines will not free up capacity. In the end I cannot build new > machines even if there is enough space on the hypervisors. > > Regards, > Ingo > ________________________________________ > Von: Marc-Aurèle Brothier [[email protected]] > Gesendet: Mittwoch, 27. September 2017 14:22 > An: [email protected] > Betreff: Re: Free capacity calculation within ACS > > What do you see in the table vm_instance for the VMs you were expecting a > release? Do they stay in Destroyed state only and are not moving to > "Expunging" state? What do you see in the logs related to the thread named > "UserVm-Scavenger". This is the one which should do the VM cleanup. What do > you have in the logs related to CapacityManagerImpl class? > > Kind regards, > Marc-Aurèle > > > > > On 26 Sep 2017, at 22:44, Jochim, Ingo <[email protected]> wrote: > > > > Hello Marc-Aurèle, > > > > we tested with the parameter host.reservation.release.period. > > We didn't see any change in the free capacity after shutdown of > machines. Not even after this period. > > Any idea why? > > > > Regards, > > Ingo > > > > > > > > > > -----Ursprüngliche Nachricht----- > > Von: Jochim, Ingo [mailto:[email protected]] > > Gesendet: Mittwoch, 6. September 2017 09:56 > > An: '[email protected]' <[email protected]> > > Betreff: AW: Free capacity calculation within ACS > > > > Many thanks. We will check. > > > > Regards, > > Ingo > > > > -----Ursprüngliche Nachricht----- > > Von: Marc-Aurèle Brothier [mailto:[email protected]] > > Gesendet: Mittwoch, 6. September 2017 09:51 > > An: [email protected] > > Betreff: Re: Free capacity calculation within ACS > > > > Apparently in 4.2.0 > > https://github.com/apache/cloudstack/blob/f0dd5994b447a6097c52f405c7c7c5 > 4c76da9c16/setup/db/db/schema-410to420.sql > > > > On Wed, Sep 6, 2017 at 9:42 AM, Jochim, Ingo <[email protected]> > > wrote: > > > >> Hello Marc-Aurèle, > >> > >> great. Didn't know that this parameter exists. > >> Do you know in which ACS version this got introduced? > >> > >> Many thanks. > >> Ingo > >> > >> -----Ursprüngliche Nachricht----- > >> Von: Marc-Aurèle Brothier [mailto:[email protected]] > >> Gesendet: Mittwoch, 6. September 2017 09:15 > >> An: [email protected] > >> Betreff: Re: Free capacity calculation within ACS > >> > >> Hi Ingo, > >> > >> You might want to look at the release period set in your installation: > >> host.reservation.release.period. This release window time is there to > >> keep the capacity of stopped machines on a host for a certain time, > >> before releasing it, in case the machine has to start again soon > >> after. And most likely for other reason maybe in the advance > >> networking mode. So try to decrease this time window and check your > capacity after that. > >> > >> Marc-Aurèle > >> > >> On Tue, Sep 5, 2017 at 9:48 AM, Jochim, Ingo <[email protected]> > >> wrote: > >> > >>> Hello all, > >>> > >>> within our CloudStack environment we like to park a couple of large > >>> machines in powered off state. > >>> Those machines are demo machines which are needed only sometimes. > >>> Those machines will get included in the capacity. That means we > >>> cannot build new machines even if there are free ressources on the > hypervisors. > >>> We don't want to solve it with overcommitment. > >>> Is there a possibility to calculate free capacity without all > >>> powered off machines? > >>> > >>> Currently we have a dirty workaround. We created an offering with 1 > >>> core and 1MB RAM and used that for all parked machines. > >>> But this is not very nice. > >>> > >>> Any ideas or comments are welcome. > >>> Thank you. > >>> Ingo > >>> > >> > >
