Hullo Naïve question follows. Sorry. I’m trying a minimal OS install on a Virtualbox VM on a mac. I’d like to get to the point where I can launch a compute node. I’ve failed with `packstack` and I’m trying `devstack`. Should this work out of the box: ie Spin up a vm with vagrant: I’m using Centos Stream 9 to ensure that I get a current(ish) version of Python. It has 9GB RAM Ensure that SELinux and firewalls are turned off Clone devstack, cd to the directory and run `stack.sh` as user `vagrant` (this fails 1/3 of the time as some repo or other resets a connection. `stack.sh` doesn’t seem to be idempotent as reinvoking it may or may not install and run the OS environment Upload a ssh keypair through the web interface Use the web interface to launch the m1.nano flavor with Cirros image (I think that this flavor is quite new as some of the documentation refers to creating such a flavor with 64MB, whereas this one has 128MB. I did try the 64MB route [with `packstack`] and concluded that at least 96MB was needed and the documentation was wrong. I couldn’t log into launchpad.net <http://launchpad.net/> to report this ☹ At this point the launch process fails with the error message: “Build of instance 157bfa1d-7f8c-4a6c-ba3a-b02fb4f4b6a9 aborted: Failed to allocate the network(s), not rescheduling.” In the web ui Afaict, the vm has enough memory (just: it’s using a bit of swap, but more cache, so it could reclaim that). I’d expected the instance to launch, and I can well believe that I’ve missed something, but the documentation seems to point all over the place for various logs. Should this approach work? Is there an alternative that’s better (e.g. use Ubuntu: I’m not keen on apt/dpkg/.deb based distros as I’ve been tripped up in the past over the dependency handling and systemd integration, so I’ve avoided this, but I can see that Canonical is spending money on OS. But so is IBM/Redhat)? Where can I find info on how to trouble shoot the failing process? tia Tim
Hi Tim, This should work fine. You will want a localrc/local.conf file to configure devstack. I didn't see that mentioned in your steps. See this section in the docs: https://docs.openstack.org/devstack/latest/#create-a-local-conf The only caveat I would mention is the VM instances in Nova will run super slow on virtualbox as it lacks the required "nested virtualization" support and will run them all in software emulation. To find the root cause of the issue in nova, I would look through the devstack@n-cpu log file (journal -u devstack@n-cpui) and the devstack@n-sch logs. Also, you might have a look at one of the nova test localrc file as an example: https://storage.gra.cloud.ovh.net/v1/AUTH_dcaab5e32b234d56b626f72581e3644c/z... Michael On Fri, May 27, 2022 at 4:16 AM <tim+openstack.org@coote.org> wrote:
Hullo
Naïve question follows. Sorry.
I’m trying a minimal OS install on a Virtualbox VM on a mac. I’d like to get to the point where I can launch a compute node. I’ve failed with `packstack` and I’m trying `devstack`.
Should this work out of the box: ie
Spin up a vm with vagrant: I’m using Centos Stream 9 to ensure that I get a current(ish) version of Python. It has 9GB RAM Ensure that SELinux and firewalls are turned off Clone devstack, cd to the directory and run `stack.sh` as user `vagrant` (this fails 1/3 of the time as some repo or other resets a connection. `stack.sh` doesn’t seem to be idempotent as reinvoking it may or may not install and run the OS environment Upload a ssh keypair through the web interface Use the web interface to launch the m1.nano flavor with Cirros image (I think that this flavor is quite new as some of the documentation refers to creating such a flavor with 64MB, whereas this one has 128MB. I did try the 64MB route [with `packstack`] and concluded that at least 96MB was needed and the documentation was wrong. I couldn’t log into launchpad.net to report this ☹ At this point the launch process fails with the error message: “Build of instance 157bfa1d-7f8c-4a6c-ba3a-b02fb4f4b6a9 aborted: Failed to allocate the network(s), not rescheduling.” In the web ui
Afaict, the vm has enough memory (just: it’s using a bit of swap, but more cache, so it could reclaim that). I’d expected the instance to launch, and I can well believe that I’ve missed something, but the documentation seems to point all over the place for various logs.
Should this approach work? Is there an alternative that’s better (e.g. use Ubuntu: I’m not keen on apt/dpkg/.deb based distros as I’ve been tripped up in the past over the dependency handling and systemd integration, so I’ve avoided this, but I can see that Canonical is spending money on OS. But so is IBM/Redhat)? Where can I find info on how to trouble shoot the failing process?
tia Tim
Thanks, Michael. Very reassuring. I’ll have a look and comment back.
On 27 May 2022, at 20:11, Michael Johnson <johnsomor@gmail.com> wrote:
Hi Tim,
This should work fine. You will want a localrc/local.conf file to configure devstack. I didn't see that mentioned in your steps. See this section in the docs: https://docs.openstack.org/devstack/latest/#create-a-local-conf
The only caveat I would mention is the VM instances in Nova will run super slow on virtualbox as it lacks the required "nested virtualization" support and will run them all in software emulation.
To find the root cause of the issue in nova, I would look through the devstack@n-cpu log file (journal -u devstack@n-cpui) and the devstack@n-sch logs.
Also, you might have a look at one of the nova test localrc file as an example: https://storage.gra.cloud.ovh.net/v1/AUTH_dcaab5e32b234d56b626f72581e3644c/z...
Michael
On Fri, May 27, 2022 at 4:16 AM <tim+openstack.org@coote.org> wrote:
Hullo
Naïve question follows. Sorry.
I’m trying a minimal OS install on a Virtualbox VM on a mac. I’d like to get to the point where I can launch a compute node. I’ve failed with `packstack` and I’m trying `devstack`.
Should this work out of the box: ie
Spin up a vm with vagrant: I’m using Centos Stream 9 to ensure that I get a current(ish) version of Python. It has 9GB RAM Ensure that SELinux and firewalls are turned off Clone devstack, cd to the directory and run `stack.sh` as user `vagrant` (this fails 1/3 of the time as some repo or other resets a connection. `stack.sh` doesn’t seem to be idempotent as reinvoking it may or may not install and run the OS environment Upload a ssh keypair through the web interface Use the web interface to launch the m1.nano flavor with Cirros image (I think that this flavor is quite new as some of the documentation refers to creating such a flavor with 64MB, whereas this one has 128MB. I did try the 64MB route [with `packstack`] and concluded that at least 96MB was needed and the documentation was wrong. I couldn’t log into launchpad.net to report this ☹ At this point the launch process fails with the error message: “Build of instance 157bfa1d-7f8c-4a6c-ba3a-b02fb4f4b6a9 aborted: Failed to allocate the network(s), not rescheduling.” In the web ui
Afaict, the vm has enough memory (just: it’s using a bit of swap, but more cache, so it could reclaim that). I’d expected the instance to launch, and I can well believe that I’ve missed something, but the documentation seems to point all over the place for various logs.
Should this approach work? Is there an alternative that’s better (e.g. use Ubuntu: I’m not keen on apt/dpkg/.deb based distros as I’ve been tripped up in the past over the dependency handling and systemd integration, so I’ve avoided this, but I can see that Canonical is spending money on OS. But so is IBM/Redhat)? Where can I find info on how to trouble shoot the failing process?
tia Tim
On Mon, 2022-05-30 at 12:06 +0100, tim+openstack.org@coote.org wrote:
Thanks, Michael. Very reassuring. I’ll have a look and comment back.
if you have an m1 mac i woudl suggest using utm with the ubuntu 20.04 image https://mac.getutm.app/gallery/ubuntu-20-04 to create a qemu vm whihc will use macos's hypervirio api to hardware acclearte the l1 vm. the l2 vms will still use qemu but by using arm based images for the host os you can get pretty good perforamce in teh vm and spin up arm based cirrous iamge with nova and get ok performance. it will be slower then nested virt but the apple silicon macs dont support that a the hardware level. i have mostly for my own use being developing https://github.com/SeanMooney/ansible_role_devstack im probably goign to rename that to "ard" by the way in the future if that link does nto work later. this repo currently has ansible playbooks that will use the upstream devstack roles we use in ci to deploy multi node devstack it can create vms using molecule/vagrant and then run the devstack install. so on a linux laptop you can do bootstrap-repo.sh molecule create molecule converge and that will create a 2 node openstack based on centos 9 stream with master devstack installed and deployed. currently the molecule role creates two 8gb vms but i have an example of using it to deploy onto externally provisioned host as a pr https://github.com/SeanMooney/ansible_role_devstack/pull/4/files if you continue to have troble deploying devstack by hand perhaps that will be of use to you. it not really ready for primetime/general use but if people find it intersting ye are welcome to fork and use as ye see fit. the molecule template seamed to set the licence to "BSD" by default. i was plaaing ot make it apache but i guess bsd is just as good so i shoudl fix that. anyway i just wanted to point out that UTM is proably beter the virtual box form a performace point of view.
On 27 May 2022, at 20:11, Michael Johnson <johnsomor@gmail.com> wrote:
Hi Tim,
This should work fine. You will want a localrc/local.conf file to configure devstack. I didn't see that mentioned in your steps. See this section in the docs: https://docs.openstack.org/devstack/latest/#create-a-local-conf
The only caveat I would mention is the VM instances in Nova will run super slow on virtualbox as it lacks the required "nested virtualization" support and will run them all in software emulation.
To find the root cause of the issue in nova, I would look through the devstack@n-cpu log file (journal -u devstack@n-cpui) and the devstack@n-sch logs.
Also, you might have a look at one of the nova test localrc file as an example: https://storage.gra.cloud.ovh.net/v1/AUTH_dcaab5e32b234d56b626f72581e3644c/z...
Michael
On Fri, May 27, 2022 at 4:16 AM <tim+openstack.org@coote.org> wrote:
Hullo
Naïve question follows. Sorry.
I’m trying a minimal OS install on a Virtualbox VM on a mac. I’d like to get to the point where I can launch a compute node. I’ve failed with `packstack` and I’m trying `devstack`.
Should this work out of the box: ie
Spin up a vm with vagrant: I’m using Centos Stream 9 to ensure that I get a current(ish) version of Python. It has 9GB RAM Ensure that SELinux and firewalls are turned off Clone devstack, cd to the directory and run `stack.sh` as user `vagrant` (this fails 1/3 of the time as some repo or other resets a connection. `stack.sh` doesn’t seem to be idempotent as reinvoking it may or may not install and run the OS environment Upload a ssh keypair through the web interface Use the web interface to launch the m1.nano flavor with Cirros image (I think that this flavor is quite new as some of the documentation refers to creating such a flavor with 64MB, whereas this one has 128MB. I did try the 64MB route [with `packstack`] and concluded that at least 96MB was needed and the documentation was wrong. I couldn’t log into launchpad.net to report this ☹ At this point the launch process fails with the error message: “Build of instance 157bfa1d-7f8c-4a6c-ba3a-b02fb4f4b6a9 aborted: Failed to allocate the network(s), not rescheduling.” In the web ui
Afaict, the vm has enough memory (just: it’s using a bit of swap, but more cache, so it could reclaim that). I’d expected the instance to launch, and I can well believe that I’ve missed something, but the documentation seems to point all over the place for various logs.
Should this approach work? Is there an alternative that’s better (e.g. use Ubuntu: I’m not keen on apt/dpkg/.deb based distros as I’ve been tripped up in the past over the dependency handling and systemd integration, so I’ve avoided this, but I can see that Canonical is spending money on OS. But so is IBM/Redhat)? Where can I find info on how to trouble shoot the failing process?
tia Tim
Thanks, Sean. Unfortunately, this is a small (16GB x86). I did respond with more error messages, but I think they’re probably too obtuse.
On 30 May 2022, at 12:53, Sean Mooney <smooney@redhat.com> wrote:
On Mon, 2022-05-30 at 12:06 +0100, tim+openstack.org@coote.org wrote:
Thanks, Michael. Very reassuring. I’ll have a look and comment back.
if you have an m1 mac i woudl suggest using utm with the ubuntu 20.04 image https://mac.getutm.app/gallery/ubuntu-20-04 to create a qemu vm whihc will use macos's hypervirio api to hardware acclearte the l1 vm. the l2 vms will still use qemu but by using arm based images for the host os you can get pretty good perforamce in teh vm and spin up arm based cirrous iamge with nova and get ok performance. it will be slower then nested virt but the apple silicon macs dont support that a the hardware level.
i have mostly for my own use being developing https://github.com/SeanMooney/ansible_role_devstack im probably goign to rename that to "ard" by the way in the future if that link does nto work later.
this repo currently has ansible playbooks that will use the upstream devstack roles we use in ci to deploy multi node devstack it can create vms using molecule/vagrant and then run the devstack install.
so on a linux laptop you can do
bootstrap-repo.sh molecule create molecule converge and that will create a 2 node openstack based on centos 9 stream with master devstack installed and deployed.
currently the molecule role creates two 8gb vms but i have an example of using it to deploy onto externally provisioned host as a pr https://github.com/SeanMooney/ansible_role_devstack/pull/4/files
if you continue to have troble deploying devstack by hand perhaps that will be of use to you.
it not really ready for primetime/general use but if people find it intersting ye are welcome to fork and use as ye see fit. the molecule template seamed to set the licence to "BSD" by default. i was plaaing ot make it apache but i guess bsd is just as good so i shoudl fix that. anyway i just wanted to point out that UTM is proably beter the virtual box form a performace point of view.
On 27 May 2022, at 20:11, Michael Johnson <johnsomor@gmail.com> wrote:
Hi Tim,
This should work fine. You will want a localrc/local.conf file to configure devstack. I didn't see that mentioned in your steps. See this section in the docs: https://docs.openstack.org/devstack/latest/#create-a-local-conf
The only caveat I would mention is the VM instances in Nova will run super slow on virtualbox as it lacks the required "nested virtualization" support and will run them all in software emulation.
To find the root cause of the issue in nova, I would look through the devstack@n-cpu log file (journal -u devstack@n-cpui) and the devstack@n-sch logs.
Also, you might have a look at one of the nova test localrc file as an example: https://storage.gra.cloud.ovh.net/v1/AUTH_dcaab5e32b234d56b626f72581e3644c/z...
Michael
On Fri, May 27, 2022 at 4:16 AM <tim+openstack.org@coote.org> wrote:
Hullo
Naïve question follows. Sorry.
I’m trying a minimal OS install on a Virtualbox VM on a mac. I’d like to get to the point where I can launch a compute node. I’ve failed with `packstack` and I’m trying `devstack`.
Should this work out of the box: ie
Spin up a vm with vagrant: I’m using Centos Stream 9 to ensure that I get a current(ish) version of Python. It has 9GB RAM Ensure that SELinux and firewalls are turned off Clone devstack, cd to the directory and run `stack.sh` as user `vagrant` (this fails 1/3 of the time as some repo or other resets a connection. `stack.sh` doesn’t seem to be idempotent as reinvoking it may or may not install and run the OS environment Upload a ssh keypair through the web interface Use the web interface to launch the m1.nano flavor with Cirros image (I think that this flavor is quite new as some of the documentation refers to creating such a flavor with 64MB, whereas this one has 128MB. I did try the 64MB route [with `packstack`] and concluded that at least 96MB was needed and the documentation was wrong. I couldn’t log into launchpad.net to report this ☹ At this point the launch process fails with the error message: “Build of instance 157bfa1d-7f8c-4a6c-ba3a-b02fb4f4b6a9 aborted: Failed to allocate the network(s), not rescheduling.” In the web ui
Afaict, the vm has enough memory (just: it’s using a bit of swap, but more cache, so it could reclaim that). I’d expected the instance to launch, and I can well believe that I’ve missed something, but the documentation seems to point all over the place for various logs.
Should this approach work? Is there an alternative that’s better (e.g. use Ubuntu: I’m not keen on apt/dpkg/.deb based distros as I’ve been tripped up in the past over the dependency handling and systemd integration, so I’ve avoided this, but I can see that Canonical is spending money on OS. But so is IBM/Redhat)? Where can I find info on how to trouble shoot the failing process?
tia Tim
Hullo again.
On 27 May 2022, at 20:11, Michael Johnson <johnsomor@gmail.com> wrote:
Hi Tim,
This should work fine. You will want a localrc/local.conf file to configure devstack. I didn't see that mentioned in your steps. See this section in the docs: https://docs.openstack.org/devstack/latest/#create-a-local-conf Sorry, I missed that step in the description. local.conf is being created.
The only caveat I would mention is the VM instances in Nova will run super slow on virtualbox as it lacks the required "nested virtualization" support and will run them all in software emulation. I’m not worried about this atm. Once I’ve got something working, I can sort that out.
To find the root cause of the issue in nova, I would look through the devstack@n-cpu log file (journal -u devstack@n-cpui) and the devstack@n-sch logs. Neither of these logs contain errors. However, if I look for ‘ERROR’ in `sudo journalctl`, I do get ~400 keystone errors, where users, domains, roles and services cannot be found. I’m not sure whether this is expected or not. They occur at the beginning of the log and may be significant, or just an artefact of the `stack.sh` script. Here’s a few (filtered by `sudo journalctl |grep keyston |grep exception|grep "Could not find"|grep -v "None admin") examples: May 30 15:57:39 node1.example.dd devstack@keystone.service[94392]: ERROR keystone.server.flask.application keystone.exception.ProjectNotFound: Could not find project: service. May 30 15:57:41 node1.example.dd devstack@keystone.service[94392]: ERROR keystone.server.flask.application keystone.exception.ServiceNotFound: Could not find service: glance. May 30 15:57:41 node1.example.dd devstack@keystone.service[94392]: ERROR keystone.server.flask.application keystone.exception.ServiceNotFound: Could not find service: glance. May 30 15:57:42 node1.example.dd devstack@keystone.service[94393]: ERROR keystone.server.flask.application keystone.exception.UserNotFound: Could not find user: cinder.
Subsequently, this is what I suspect is the smoking gun, as it precedes the timeout: May 30 16:14:45 node1.example.dd nova-compute[133470]: ERROR os_brick.initiator.connectors.iscsi Command: iscsiadm -m discoverydb -o show -P 1 May 30 16:14:45 node1.example.dd nova-compute[133470]: ERROR os_brick.initiator.connectors.iscsi Exit code: 21 May 30 16:14:45 node1.example.dd nova-compute[133470]: ERROR os_brick.initiator.connectors.iscsi Stdout: 'SENDTARGETS:\nNo targets found.\niSNS:\nNo targets found.\nSTATIC:\nNo targets found.\nFIRMWARE:\nNo targets found.\n' I have no idea what that means. My guess is that there’s supposed to be a fake iscsi device and it’s not on a local network, but that’s really just a guess.
Also, you might have a look at one of the nova test localrc file as an example: https://storage.gra.cloud.ovh.net/v1/AUTH_dcaab5e32b234d56b626f72581e3644c/z...
Presumably, this example would need things like the various IP addresses changing to be useful. What would one then do with it?
Michael
Tim
participants (3)
-
Michael Johnson
-
Sean Mooney
-
tim+openstack.org＠coote.org