Age | Commit message (Collapse) | Author | Files | Lines |
|
|
|
The per port model potentially requires an increase in memory
resource requirements (which is limited by labs) to support the
same number of ports and configuration as the shared port model.
Set linux:network:openvswitch:per_port_memory explicitly to true
to enable per port mempools support for DPDK devices.
Change-Id: I130885afc50e7a047f8835113d370840827ad718
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
Change-Id: Id49f26a2615e2fc06e94eeaf2e9200e83625e6c9
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
Deploy the OpenStack API services based on roles to
prevent issues with absent database tables since db_sync
runs only on the nodes with primary role.
Change-Id: I04cf3ce0dd59afd93b8a0dfcf060fbd7e7411c82
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
Previously we only synced the scripts subdir, but going forward
we will need the full contents of the IEC repo on all cluster nodes.
Change-Id: I88edd4885875048d50d28c1eac9fd413dc2b6ffb
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
Executing deploy.sh multiple times led to duplicating the ip rules.
Change-Id: Iad5886a851970f166996226fa3d115a93113c6db
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
To bypass Docker 'bridge'-backed network isolation, we previously
added an extra routing hop, which broke access from inside the
'mcpcontrol' Docker network (typically 10.20.0.0/24) to its
bridge address (10.20.0.1), leading to DNS issues on Salt Master.
This change leverages policy based routing to only add the extra
routing hop for connections originating from the default Docker
bridge network ('docker0'). Note that other Docker networks
using the 'bridge' driver are still isolated from 'mcpcontrol'.
Fixes: d9b44acb
Change-Id: Ib92901c3278ae9b815f28f26d4c26f82bcadacd6
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
|
|
Optimized for LF-POD2 as nic assigned to private/dpdk interface
and pinned cores resides on numa #0. Core #11 is for DPDK,
the rest four cores for PMDs.
Change-Id: Icca701bc1a66f3672b8511e0245c82ca29788a8b
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
Set timeout value for snat punts to zero to turn
off the rate limiting and installation of learnt flows.
Change-Id: I79dad8fd0f925bfc11d7dc1678c3a414dc35fa56
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
|
|
Recent virsh/Docker network rework changed mcpcontrol (previously
a virsh-managed network) into a Docker-controlled network using
the 'bridge' driver.
As a consequence, Docker now isolates traffic from 'mcpcontrol'
network from the default Docker bridge network ('docker0') using
iptables rules that check input/output interfaces.
Yardstick (and any other Docker container hooked via 'docker0')
will not be able to ssh into Salt master due to this isolation.
One possible workaround would be to explicitly ACCEPT traffic
from 'docker0' going to Salt master. However, this is only
properly supported starting with Docker 17.06, while most CI hosts
and end users are still using 17.05 or older.
In older Docker releases, DOCKER-USER iptables table was not
avaiable, so injecting custom iptables and making them persistent
is not only complicated, it's also prone to subtle errors.
Another way to bypass the iptables rules is to route the packets
coming from our new Docker network via another bridge before
letting them find their way into 'docker0'.
This change adds a new route for the Salt master host (note that
MaaS container will not benefit from this) via the PXE bridge on
the jumphost (which can be either a real Linux bridge for baremetal
deployments or a virsh-managed network); adding one extra network
hop for each packet going between our 'mcpcontrol' Docker network
and 'docker0', effectively bypassing the Docker-enforced iptables
DROP.
Change-Id: Id8ac7a638c778887b361c9b64c320664c88f59fd
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
* update system reclass
* rectify telemetry redis options
Change-Id: I6dca1ae52e7f7d73a90e53fceddca8e86872651b
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
|
|
|
|
Change-Id: I791436f512dea6c6bc61133c4122ac872950af8e
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
Upstream change [1] switched from old qemu-nbd preseeding of VCP VMs
to using a cloud-init + configuration drive. This breaks on AArch64
with "IDE controllers are unsupported for this QEMU binary or machine
type", so switch back to using qemu-nbd.
[1] https://github.com/Mirantis/reclass-system-salt-model/commit/c0e4807
Change-Id: I0dfeb638d408343c76a73fafa503048a79ce1f6e
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
NOTE: only os-nosdn-nofeature-noha is parameterized for now.
- move config drive & disk creation from prepare_vms to create_vms;
- make default disk size(s) configurable based on scenario defaults
and vPDF;
* compute nodes require 2 disks to be defined in vPDF, since the
pillar reclass model assumes /dev/vdb is reserved for cinder;
* if multiple disks are defined in vPDF, they are created and
attached accordinly (only ctl01 and cmp nodes are parameterized
in this change; only for the os-nosdn-nofeature-noha scenario);
- vCPU specifications are deduced based on vPDF (sockets, cores);
* threads/core is hard set to 2 since vPDF does not have a key
for it;
* NUMA resources are distributed evenly based on the number of
sockets configured in PDF;
* no less than the mininum requirement for a scenario is allocated
(e.g. if PDF specifies 2 cores, but the scenario requires at
least 4 cores, the larger value will be used);
- RAM is deduced based on PDF (but no less than the mininum req is
allocated, e.g. if PDF specifies 2GB RAM for computes, but the
scenario requires at least 8GB, the larger value will be used);
Change-Id: I97188aa2a1006865b8429eb6483e10c76795f7d2
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
There is no enough memory (default 4k pages) for services
like libvirt, which cannot fork child processes.
Change-Id: I44d8efd7cafb52a7c823c02738c1d321017aa7a3
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
Required only for Rally validation in cinder scenarios,
there is no useful functionaly in terms of cluster.
Change-Id: Idc4d62cbbc9974972e9d492b5a419342077e3d9a
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
Sometimes instance doesn't get ip address from dhcp server, which
resides only on gateway node, so run additional dhcp/metadata agents
on compute nodes to handle tenant networks in place.
Change-Id: If1d74af665cf8db64b09f846fac7192f76abdb25
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
The per port memory model provides a more transparent memory usage model
and avoids pool exhaustion due to competing memory requirements for
interfaces. (http://docs.openvswitch.org/en/latest/topics/dpdk/memory/)
Change-Id: I5add0f49cdcdf2fc3d24affee10a275abe3ca46a
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
- bump Pharos git submodule to allow PODs with fewer nodes;
- add `k8-calico-iec-noha` scenario definition for Akraino
IEC basic configuration;
- add `k8-calico-iec-vcp-noha` scenario definition for Akraino
IEC nested (virtualized control plane) configuration;
- add `akraino_iec` state, which will leverage the Akraino IEC
bootstrap scripts from [1];
- replace system.reboot salt call with cmd.run 'reboot' as it's more
reliable;
- use kernel 4.15 for AArch64 K8 IEC scenarios;
NOTE: These scenarios will not be released in OPNFV since don't rely
on Salt formulas but instead of Akraino IEC scripts to install K8s.
[1] https://gerrit.akraino.org/r/#/q/project:iec
Change-Id: I4e538e0563d724cd3fd5c4d462ddc22d0c739402
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
Change-Id: I2b41ce2e275bb053fa2590654ea7fa432b0c857f
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
* add opendaylight password (removed from system level)
* get updated ovn system class w/o mysql settings
* enable ceilometer user back (removed along with outdated service/endpoints)
* adjsut check interval of haproxy for noha scenarios since there is
only one backend for services, i.e. failover ain't expected
Change-Id: Iedee290e1cfcf838998bd44dc09a729d143974ac
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
|
|
After Rocky support was added upstream to salt-formula-neutron, our
FDIO patch continued to be applied only for Queens, so refresh the
patch by switching to Rocky.
Change-Id: If0bbb9c4ec674d386ceade00ef8fe936482fb49c
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
Change-Id: I745a838b1f2f294b6c455700509ddf4b0264446f
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
This reverts commit ac56d7b14f46b05f497b3dca4b6a4b0bfedd83e2.
The original patch has been merged (https://review.openstack.org/643011)
Change-Id: I3a7cd825f371e375d36256143b4b8c91f90ee26e
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
Certain kernels (e.g. 4.4.0-101+ in Ubuntu) no longer automatically
ack the partition table update after `kpartx -a /dev/nbdX`, see [1].
To avoid another dependency on `parted` packages, use `partx` from
`util-linux`, which is already installed as a dependency of e2fsprogs.
[1] https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1743026
Change-Id: Ibd993fe210c1a11814e89a66759568d4d117d613
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
* update gnocchi to 4.3
* remove outdated ceilometer api
Change-Id: I7adaf3ddc76d93531b6b0997b684672b80f2992f
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
Create 2 systemd services on the jumphost that will handle veth
pairs creation, respectively adding them to virsh/real bridges.
This allows us to set docker containers restart policy to 'always',
enabling persistent Salt Master/MaaS containers across jumphost
reboots.
NOTE: libvirt creates virtual networks async, hence the need for
retrying hooking veths to them.
JIRA: FUEL-406
Change-Id: I1ca033cb5eb854b577b57bb2387a58bd9605a5bb
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
Change-Id: Id75ffe4db808a4ec250ba8b86c5d49f1206c3784
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
Also re-align resources for virtual scenarios.
Change-Id: Id0d55407fd5b1720a24e30c364219f8b08e89d06
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
Bug: https://bugs.launchpad.net/nova/+bug/1809123
Change-Id: I14622c21826aeeddac6ea7bf7f9d116cd3e68cfb
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
|
|
As reported in [1], kernel 4.4 seems to break nested virtualization,
add a fatal check against it.
[1] https://bugs.launchpad.net/ubuntu/+source/qemu/+bug/1797332
Change-Id: I0aef8a7340dd82bfeb2e58c9642623b9ec13dca5
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
Some PODs are fast enough to get past installing, syncing and using
MaaS to provision the OS on the baremetal nodes before the 1h mine
refresh.
Since mine.update operation is fast enough to go unnoticed and we
only collect IP addresses, grains and pem entries, schedule it every
15 minutes.
Due to reclass class inheritance, we can't easily override this via
pillar data, so handle it via entrypoint.sh.
Change-Id: I0d8ed2da838ad09c94e9327d0131d3e239de4f08
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
Change-Id: Ifc4fff90551344c69295990b220f0778967887a4
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
|
|
|
|
Previously, Salt Master CA mine was only sent once, during
salt.minion.ca state execution at cfg01 bringup / bootstrap.
This causes possible issues with:
- Salt Master container restart (mine data is lost);
- UNH Lab deployment (uknown rootcause, might be related to XFS and
overlay2 being used with Docker on CentOS);
To bypass this issue, make x509.get_pem_entries module send mine data
at the default mine interval (60 minutes).
Change-Id: I5f6334ae18f5af6cbe0a164791603b67f0a3668f
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
- replace mas01 VM with a Docker container;
- drop `mcpcontrol` virsh-managed network, including special handling
previously required for it across all scripts;
- drop infrastructure VMs handling from scripts, the only VMs we still
handle are cluster VMs for virtual and/or hybrid deployments;
- drop SSH server from mas01;
- stop running linux state on mas01, as all prerequisites are properly
handled durin Docker build or via entrypoint.sh - for completeness,
we still keep pillar data in sync with the actual contents of mas01
configuration, so running the state manually would still work;
- make port 5240 available on the jumpserver for MaaS dashboard access;
- docs: update diagrams and text to reflect the new changes;
Change-Id: I6d9424995e9a90c530fd7577edf401d552bab929
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
While the minions are working their jobs the CLI is waiting for the
first initial timeout period (timeout) to start. When that hits,
the CLI sends sends the first "find_job" query. This kicks off the
gather_job_timeout timer. Sometimes a minion doesn't respond to the request
within the gather_job_timeout time period (default is 10s), so rise up
this value to give a chance for a minion to report actual status.
Change-Id: Ic3756b82fdeb17718870ab30e9578263d25309f7
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
|
|
|
|
Change-Id: I3bbe3e4be520ccac198654bb4a7d493aa8450023
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
Change-Id: I7709c9ca9e701b656447154919eb084a710f49af
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
The PaxOsgi logging has a performance impact
(i.e. makes pressure to the Java GC).
Change-Id: Ic0bc2c0d1cfac195a04d1cfa90fa7fa47fc37612
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
|