Tungsten Fabric入门宝典丨关于集群更新的那些事
Posted
tags:
篇首语：本文由小常识网(cha138.com)小编为大家整理，主要介绍了Tungsten Fabric入门宝典丨关于集群更新的那些事相关的知识，希望对你有一定的参考价值。
Tungsten Fabric入门宝典系列文章，来自技术大牛倾囊相授的实践经验，由TF中文社区为您编译呈现，旨在帮助新手深入理解TF的运行、安装、集成、调试等全流程。如果您有相关经验或疑问，欢迎与我们互动，并与社区极客们进一步交流。更多TF技术文章，请点击公号底部按钮＞学习＞文章合集。
作者：Tatsuya Naganawa 译者：TF编译组
集群范围内的更新是很重要的功能，可以在保证生产集群的SLA的条件下提供集群中最新功能的有效性。
由于Tungsten Fabric在MPLS-***中使用类似的协议，因此，根据我的尝试，即使Control的模块版本和vRouter的模块版本不同，基本的互操作性还是有的。
因此，总体思路是，首先一个一个地更新controller，然后并根据需要使用vMotion或maintence模式逐一更新vRouter。
另外，Tungsten Fabric controller还支持名为ISSU的奇特功能，但是我认为这个名称有些混乱，因为Tungsten Fabric controller与route reflector非常相似，而不是routing-engine。
因此，基本思路是，首先将所有配置（configs）复制到新创建的controller（或route reflectors），然后更新vRouter设置（如果服务器可以重新启动，则更新vRouter模块）以使用这些新的controller。通过这个过程，vRouter模块更新的回滚操作也将更加容易。
下面我来描述一下这个过程。
就地更新（in-place update）
由于ansible-deployer遵循等幂行为（idempotent behavior），因此更新与安装没有太大区别。以下的命令将更新所有模块。
cd contrail-ansible-deployer
git pull
vi config/instances.yaml
(update CONTRAIL_CONTAINER_TAG)
ansible-playbook -e orchestrator=xxx -i inventory/ playbooks/install_contrail.yml一个限制是由于此命令几乎同时重启了所有节点，因此不容易做到一一重启controller和vRouter。另外，从instances.yaml中删除其它节点将不起作用，因为一个节点的更新需要其它节点的某些参数。
- 例如，vRouter更新需要控件的IP，该IP是从instances.yaml中的control role节点推导出的
为了克服这个问题，从R2005版本开始，ziu.yaml就加入了这个功能，至少对于控制平面来说，要逐一更新。
cd contrail-ansible-deployer
git pull
vi config/instances.yaml
(update CONTRAIL_CONTAINER_TAG)
ansible-playbook -e orchestrator=xxx -i inventory/ playbooks/ziu.yml
ansible-playbook -e orchestrator=xxx -i inventory/ playbooks/install_contrail.yml就我试过的情况来看，当控制平面更新时，它会进行串联更新和重启控制进程，所以没有看到掉包的现象。
-
在install_contrail.yaml期间，控制进程的重启是被跳过的，因为它们已经更新了。
- 当执行vrouter-agent重启时，仍然会出现一些丢包现象，所以如果可以的话，建议进行工作负载迁移。
ISSU
即使容器格式的差异很大（例如从4.x到5.x），我们也可以使用ISSU，因为它创建了一个新的controller集群，并在其中复制数据。
首先，我来描述最简单的情况，即一个旧controller和一个新controller，以查看整个过程。所有命令都在新的controller上键入。
old-controller:
ip: 172.31.2.209
hostname: ip-172-31-2-209
new-controller:
ip: 172.31.1.154
hostname: ip-172-31-1-154
(both controllers are installed with this instances.yaml)
provider_config:
bms:
ssh_user: root
ssh_public_key: /root/.ssh/id_rsa.pub
ssh_private_key: /root/.ssh/id_rsa
domainsuffix: local
ntpserver: 0.centos.pool.ntp.org
instances:
bms1:
provider: bms
roles:
config_database:
config:
control:
analytics:
analytics_database:
webui:
ip: x.x.x.x ## controller‘s ip
contrail_configuration:
CONTRAIL_CONTAINER_TAG: r5.1
KUBERNETES_CLUSTER_PROJECT: {}
JVM_EXTRA_OPTS: "-Xms128m -Xmx1g"
global_configuration:
CONTAINER_REGISTRY: tungstenfabric
[commands]
1. 停止批处理作业
docker stop config_devicemgr_1
docker stop config_schema_1
docker stop config_svcmonitor_1
2. 在cassandra上注册新的control，在它们之间运行bgp
docker exec -it config_api_1 bash
python /opt/contrail/utils/provision_control.py --host_name ip-172-31-1-154.local --host_ip 172.31.1.154 --api_server_ip 172.31.2.209 --api_server_port 8082 --oper add --router_asn 64512 --ibgp_auto_mesh
3. 在控制器之间同步数据
vi contrail-issu.conf
(write down this)
[DEFAULTS]
old_rabbit_address_list = 172.31.2.209
old_rabbit_port = 5673
new_rabbit_address_list = 172.31.1.154
new_rabbit_port = 5673
old_cassandra_address_list = 172.31.2.209:9161
old_zookeeper_address_list = 172.31.2.209:2181
new_cassandra_address_list = 172.31.1.154:9161
new_zookeeper_address_list = 172.31.1.154:2181
new_api_info={"172.31.1.154": [("root"), ("password")]} ## ssh public-key can be used
image_id=`docker images | awk ‘/config-api/{print $3}‘ | head -1`
docker run --rm -it --network host -v $(pwd)/contrail-issu.conf:/etc/contrail/contrail-issu.conf --entrypoint /bin/bash -v /root/.ssh:/root/.ssh $image_id -c "/usr/bin/contrail-issu-pre-sync -c /etc/contrail/contrail-issu.conf"
4. 启动进程进行实时数据同步
docker run --rm --detach -it --network host -v $(pwd)/contrail-issu.conf:/etc/contrail/contrail-issu.conf --entrypoint /bin/bash -v /root/.ssh:/root/.ssh --name issu-run-sync $image_id -c "/usr/bin/contrail-issu-run-sync -c /etc/contrail/contrail-issu.conf"
(如果需要，请检查日志)
docker exec -t issu-run-sync tail -f /var/log/contrail/issu_contrail_run_sync.log
5. (更新 vrouters)
6. 在结束后停止作业，同步所有数据
docker rm -f issu-run-sync
image_id=`docker images | awk ‘/config-api/{print $3}‘ | head -1`
docker run --rm -it --network host -v $(pwd)/contrail-issu.conf:/etc/contrail/contrail-issu.conf --entrypoint /bin/bash -v /root/.ssh:/root/.ssh --name issu-run-sync $image_id -c "/usr/bin/contrail-issu-post-sync -c /etc/contrail/contrail-issu.conf"
docker run --rm -it --network host -v $(pwd)/contrail-issu.conf:/etc/contrail/contrail-issu.conf --entrypoint /bin/bash -v /root/.ssh:/root/.ssh --name issu-run-sync $image_id -c "/usr/bin/contrail-issu-zk-sync -c /etc/contrail/contrail-issu.conf"
7. 从cassandra中去掉旧的节点，加入新的节点
vi issu.conf
(write down this)
[DEFAULTS]
db_host_info={"172.31.1.154": "ip-172-31-1-154.local"}
config_host_info={"172.31.1.154": "ip-172-31-1-154.local"}
analytics_host_info={"172.31.1.154": "ip-172-31-1-154.local"}
control_host_info={"172.31.1.154": "ip-172-31-1-154.local"}
api_server_ip=172.31.1.154
docker cp issu.conf config_api_1:issu.conf
docker exec -it config_api_1 python /opt/contrail/utils/provision_issu.py -c issu.conf
8. 启动批处理作业
docker start config_devicemgr_1
docker start config_schema_1
docker start config_svcmonitor_1以下将是可能的检查点。
-
步骤3之后，你可以尝试使用contrail-api-cli ls -l *来查看所有数据是否已成功复制，并且可以通过ist.py ctr nei来查看controller之间的ibgp是否已启动。
- 步骤4之后，你可以修改旧数据库，以查看更改是否可以成功传播到新数据库。
接下来，我将在使用编排器和两个vRouter的条件下，讨论更为实际的情况。
编排器集成
为了说明结合编排器的情况，我尝试用ansible-deployer部署两个vRouters和kubernetes。
即使与编排器结合使用，总体过程也不会有太大不同。
需要注意的是，需要在什么时候将kube-manager更改为新的。
从某种意义上讲，由于kube-manager动态订阅了来自kube-apiserver的事件并更新了Tungsten Fabric配置数据库（config-database），因此它类似于批处理作业，例如schema-transformer、svc-monitor和device-manager。因此，我使用此类批处理作业，同时停止并启动了旧的或新的kube-manager（实际上也包括webui），但是可能需要根据每个设置进行更改。
这个示例中的总体过程如下所示。
1.设置一个controller（带有一个kube-manager和kubernetes-master）和两个vRouter
2.设置一个新的controller（带有一个kube-manager，但kubernetes-master与旧的controller相同）
3.停止批处理作业、新controller的kube-manager和webui
4.启动ISSU进程并继续执行，直到开始运行同步（run-sync）
->将在controller之间建立iBGP
5.根据新controller的ansible-ployer一个一个地更新vRouter
->将一个vRouter移至新的vRouter时，新的controller也将获得k8s-default-pod-network的路由目标（route-target），并且容器之间仍然可以ping通正常工作（ist.py ctr路由摘要以及ping的结果将在稍后附上）
6.将所有vRouter移至新的controller后，停止批处理作业、旧controller上的kube-manager和webui
之后，继续执行ISSU进程，新controller上启动批处理作业， kube-manager和webui
->这个阶段从开始到结束，你都无法手动更改config-database，因此可能需要一些维护时间
（整个过程可能会持续5至15分钟，可以ping通，但是直到启动新的kube-manager时，新容器的创建才能正常工作）
7.最后，停止旧的节点上的control、config和config-database更新vRouters时，我使用了controller的provider: bms-maint，k8s_master和vRouter，它们都已经更改为新的，以避免由于容器重启而造成干扰。我附上了原始instances.yaml和更新vRouter的instances.yaml，以便大家获取更多详细信息。
我还将在每个阶段附加ist.py ctr nei和ist.py ctr路由摘要的结果，以说明发生的相关细节。
- 请注意，在此示例中，我实际上并未更新模块，因为此设置主要是为了突出ISSU进程（因为即使模块版本相同，ansible-deployer也会重新创建vrouter-agent容器，同时即使完成实际的模块更新，丢包的数量也不会有太大不同。）
old-controller: 172.31.19.25
new-controller: 172.31.13.9
two-vRouters: 172.31.25.102, 172.31.33.175
开始issu之前：
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.19.25 ctr nei
Introspect Host: 172.31.19.25
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
| peer | peer_address | peer_asn | encoding | peer_type | state | send_state | flap_count | flap_time |
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
| ip-172-31-25-102.local | 172.31.25.102 | 0 | XMPP | internal | Established | in sync | 0 | n/a |
| ip-172-31-33-175.local | 172.31.33.175 | 0 | XMPP | internal | Established | in sync | 0 | n/a |
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
[root@ip-172-31-13-9 ~]#
[root@ip-172-31-13-9 ~]#
[root@ip-172-31-13-9 ~]#
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.13.9 ctr nei
Introspect Host: 172.31.13.9
[root@ip-172-31-13-9 ~]#
-> iBGP is not configured yet
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.19.25 ctr route summary
Introspect Host: 172.31.19.25
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
| name | prefixes | paths | primary_paths | secondary_paths | infeasible_paths |
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
| default-domain:default- | 0 | 0 | 0 | 0 | 0 |
| project:__link_local__:__link_local__.inet.0 | | | | | |
| default-domain:default-project:dci- | 0 | 0 | 0 | 0 | 0 |
| network:__default__.inet.0 | | | | | |
| default-domain:default-project:dci-network:dci- | 0 | 0 | 0 | 0 | 0 |
| network.inet.0 | | | | | |
| default-domain:default-project:default-virtual- | 0 | 0 | 0 | 0 | 0 |
| network:default-virtual-network.inet.0 | | | | | |
| inet.0 | 0 | 0 | 0 | 0 | 0 |
| default-domain:default-project:ip-fabric:ip- | 7 | 7 | 2 | 5 | 0 |
| fabric.inet.0 | | | | | |
| default-domain:k8s-default:k8s-default-pod-network | 7 | 7 | 4 | 3 | 0 |
| :k8s-default-pod-network.inet.0 | | | | | |
| default-domain:k8s-default:k8s-default-service- | 7 | 7 | 1 | 6 | 0 |
| network:k8s-default-service-network.inet.0 | | | | | |
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
[root@ip-172-31-13-9 ~]#
[root@ip-172-31-13-9 ~]#
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.13.9 ctr route summary
Introspect Host: 172.31.13.9
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
| name | prefixes | paths | primary_paths | secondary_paths | infeasible_paths |
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
| default-domain:default- | 0 | 0 | 0 | 0 | 0 |
| project:__link_local__:__link_local__.inet.0 | | | | | |
| default-domain:default-project:dci- | 0 | 0 | 0 | 0 | 0 |
| network:__default__.inet.0 | | | | | |
| default-domain:default-project:dci-network:dci- | 0 | 0 | 0 | 0 | 0 |
| network.inet.0 | | | | | |
| default-domain:default-project:default-virtual- | 0 | 0 | 0 | 0 | 0 |
| network:default-virtual-network.inet.0 | | | | | |
| inet.0 | 0 | 0 | 0 | 0 | 0 |
| default-domain:default-project:ip-fabric:ip- | 0 | 0 | 0 | 0 | 0 |
| fabric.inet.0 | | | | | |
| default-domain:k8s-default:k8s-default-pod-network | 0 | 0 | 0 | 0 | 0 |
| :k8s-default-pod-network.inet.0 | | | | | |
| default-domain:k8s-default:k8s-default-service- | 0 | 0 | 0 | 0 | 0 |
| network:k8s-default-service-network.inet.0 | | | | | |
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
[root@ip-172-31-13-9 ~]#
-> 在新的控制器里没有导入路由
[root@ip-172-31-19-25 contrail-ansible-deployer]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE
cirros-deployment-75c98888b9-6qmcm 1/1 Running 0 4m58s 10.47.255.249 ip-172-31-25-102.ap-northeast-1.compute.internal <none>
cirros-deployment-75c98888b9-lxq4k 1/1 Running 0 4m58s 10.47.255.250 ip-172-31-33-175.ap-northeast-1.compute.internal <none>
[root@ip-172-31-19-25 contrail-ansible-deployer]#
/ # ip -o a
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue qlen 1000 link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
1: lo inet 127.0.0.1/8 scope host lo valid_lft forever preferred_lft forever
13: eth0@if14: <BROADCAST,MULTICAST,UP,LOWER_UP,M-DOWN> mtu 1500 qdisc noqueue link/ether 02:6b:dc:98:ac:95 brd ff:ff:ff:ff:ff:ff
13: eth0 inet 10.47.255.249/12 scope global eth0 valid_lft forever preferred_lft forever
/ # ping 10.47.255.250
PING 10.47.255.250 (10.47.255.250): 56 data bytes
64 bytes from 10.47.255.250: seq=0 ttl=63 time=2.155 ms
64 bytes from 10.47.255.250: seq=1 ttl=63 time=0.904 ms
^C
--- 10.47.255.250 ping statistics ---
2 packets transmitted, 2 packets received, 0% packet loss
round-trip min/avg/max = 0.904/1.529/2.155 ms
/ #
-> 两个vRouter每个都有一个容器，在两个容器之间ping的结果正常。
在provision_control之后:
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.19.25 ctr nei
Introspect Host: 172.31.19.25
+------------------------+---------------+----------+----------+-----------+-------------+-----------------+------------+-----------+
| peer | peer_address | peer_asn | encoding | peer_type | state | send_state | flap_count | flap_time |
+------------------------+---------------+----------+----------+-----------+-------------+-----------------+------------+-----------+
| ip-172-31-13-9.local | 172.31.13.9 | 64512 | BGP | internal | Idle | not advertising | 0 | n/a |
| ip-172-31-25-102.local | 172.31.25.102 | 0 | XMPP | internal | Established | in sync | 0 | n/a |
| ip-172-31-33-175.local | 172.31.33.175 | 0 | XMPP | internal | Established | in sync | 0 | n/a |
+------------------------+---------------+----------+----------+-----------+-------------+-----------------+------------+-----------+
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.13.9 ctr nei
Introspect Host: 172.31.13.9
[root@ip-172-31-13-9 ~]#
-> iBGP 在老的控制器上，但是新的控制器还没有那些配置（在执行pre-sync后，这会被复制到新的控制器上）
在run-sync之后:
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.19.25 ctr nei
Introspect Host: 172.31.19.25
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
| peer | peer_address | peer_asn | encoding | peer_type | state | send_state | flap_count | flap_time |
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
| ip-172-31-13-9.local | 172.31.13.9 | 64512 | BGP | internal | Established | in sync | 0 | n/a |
| ip-172-31-25-102.local | 172.31.25.102 | 0 | XMPP | internal | Established | in sync | 0 | n/a |
| ip-172-31-33-175.local | 172.31.33.175 | 0 | XMPP | internal | Established | in sync | 0 | n/a |
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.13.9 ctr nei
Introspect Host: 172.31.13.9
+-----------------------+--------------+----------+----------+-----------+-------------+------------+------------+-----------+
| peer | peer_address | peer_asn | encoding | peer_type | state | send_state | flap_count | flap_time |
+-----------------------+--------------+----------+----------+-----------+-------------+------------+------------+-----------+
| ip-172-31-19-25.local | 172.31.19.25 | 64512 | BGP | internal | Established | in sync | 0 | n/a |
+-----------------------+--------------+----------+----------+-----------+-------------+------------+------------+-----------+
[root@ip-172-31-13-9 ~]#
-> iBGP 建立起来了，ctr route summary没有改变，因为新的控制器没有k8s-default-pod-network的路由目标（route-target），路由目标（route target）过滤组织了导入这些前缀。
在迁移节点到新的controller之后:
/ # ping 10.47.255.250
PING 10.47.255.250 (10.47.255.250): 56 data bytes
64 bytes from 10.47.255.250: seq=0 ttl=63 time=1.684 ms
64 bytes from 10.47.255.250: seq=1 ttl=63 time=0.835 ms
64 bytes from 10.47.255.250: seq=2 ttl=63 time=0.836 ms
(snip)
64 bytes from 10.47.255.250: seq=37 ttl=63 time=0.878 ms
64 bytes from 10.47.255.250: seq=38 ttl=63 time=0.823 ms
64 bytes from 10.47.255.250: seq=39 ttl=63 time=0.820 ms
64 bytes from 10.47.255.250: seq=40 ttl=63 time=1.364 ms
64 bytes from 10.47.255.250: seq=44 ttl=63 time=2.209 ms
64 bytes from 10.47.255.250: seq=45 ttl=63 time=0.869 ms
64 bytes from 10.47.255.250: seq=46 ttl=63 time=0.857 ms
64 bytes from 10.47.255.250: seq=47 ttl=63 time=0.855 ms
64 bytes from 10.47.255.250: seq=48 ttl=63 time=0.845 ms
64 bytes from 10.47.255.250: seq=49 ttl=63 time=0.842 ms
64 bytes from 10.47.255.250: seq=50 ttl=63 time=0.885 ms
64 bytes from 10.47.255.250: seq=51 ttl=63 time=0.891 ms
64 bytes from 10.47.255.250: seq=52 ttl=63 time=0.909 ms
64 bytes from 10.47.255.250: seq=53 ttl=63 time=0.867 ms
64 bytes from 10.47.255.250: seq=54 ttl=63 time=0.884 ms
64 bytes from 10.47.255.250: seq=55 ttl=63 time=0.865 ms
64 bytes from 10.47.255.250: seq=56 ttl=63 time=0.840 ms
64 bytes from 10.47.255.250: seq=57 ttl=63 time=0.877 ms
^C
--- 10.47.255.250 ping statistics ---
58 packets transmitted, 55 packets received, 5% packet loss
round-trip min/avg/max = 0.810/0.930/2.209 ms
/ #
-> 在vrouter-agent 重启后，可以看到丢了3个包(序号40-44)。在迁移vRouter到新的之后，ping工作地很好。
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.19.25 ctr nei
检查主机：172.31.19.25
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
| peer | peer_address | peer_asn | encoding | peer_type | state | send_state | flap_count | flap_time |
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
| ip-172-31-13-9.local | 172.31.13.9 | 64512 | BGP | internal | Established | in sync | 0 | n/a |
| ip-172-31-33-175.local | 172.31.33.175 | 0 | XMPP | internal | Established | in sync | 0 | n/a |
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.13.9 ctr nei
Introspect Host: 172.31.13.9
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
| peer | peer_address | peer_asn | encoding | peer_type | state | send_state | flap_count | flap_time |
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
| ip-172-31-19-25.local | 172.31.19.25 | 64512 | BGP | internal | Established | in sync | 0 | n/a |
| ip-172-31-25-102.local | 172.31.25.102 | 0 | XMPP | internal | Established | in sync | 0 | n/a |
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
[root@ip-172-31-13-9 ~]#
-> 两个控制器具有XMPP连接，建立了IBGP
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.19.25 ctr route summary
检查主机: 172.31.19.25
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
| name | prefixes | paths | primary_paths | secondary_paths | infeasible_paths |
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
| default-domain:default- | 0 | 0 | 0 | 0 | 0 |
| project:__link_local__:__link_local__.inet.0 | | | | | |
| default-domain:default-project:dci- | 0 | 0 | 0 | 0 | 0 |
| network:__default__.inet.0 | | | | | |
| default-domain:default-project:dci-network:dci- | 0 | 0 | 0 | 0 | 0 |
| network.inet.0 | | | | | |
| default-domain:default-project:default-virtual- | 0 | 0 | 0 | 0 | 0 |
| network:default-virtual-network.inet.0 | | | | | |
| inet.0 | 0 | 0 | 0 | 0 | 0 |
| default-domain:default-project:ip-fabric:ip- | 7 | 7 | 1 | 6 | 0 |
| fabric.inet.0 | | | | | |
| default-domain:k8s-default:k8s-default-pod-network | 7 | 7 | 1 | 6 | 0 |
| :k8s-default-pod-network.inet.0 | | | | | |
| default-domain:k8s-default:k8s-default-service- | 7 | 7 | 0 | 7 | 0 |
| network:k8s-default-service-network.inet.0 | | | | | |
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.13.9 ctr route summary
检查主机: 172.31.13.9
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
| name | prefixes | paths | primary_paths | secondary_paths | infeasible_paths |
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
| default-domain:default- | 0 | 0 | 0 | 0 | 0 |
| project:__link_local__:__link_local__.inet.0 | | | | | |
| default-domain:default-project:dci- | 0 | 0 | 0 | 0 | 0 |
| network:__default__.inet.0 | | | | | |
| default-domain:default-project:dci-network:dci- | 0 | 0 | 0 | 0 | 0 |
| network.inet.0 | | | | | |
| default-domain:default-project:default-virtual- | 0 | 0 | 0 | 0 | 0 |
| network:default-virtual-network.inet.0 | | | | | |
| inet.0 | 0 | 0 | 0 | 0 | 0 |
| default-domain:default-project:ip-fabric:ip- | 7 | 7 | 1 | 6 | 0 |
| fabric.inet.0 | | | | | |
| default-domain:k8s-default:k8s-default-pod-network | 7 | 7 | 3 | 4 | 0 |
| :k8s-default-pod-network.inet.0 | | | | | |
| default-domain:k8s-default:k8s-default-service- | 7 | 7 | 1 | 6 | 0 |
| network:k8s-default-service-network.inet.0 | | | | | |
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
[root@ip-172-31-13-9 ~]#
-> 因为两个控制器具有至少一个容器来自k8s-default-pod-network, 它们使用iBGP来交换前缀，因此它们具有同一个前缀
在将第二个vrouter迁移到新的控制器之后:
/ # ping 10.47.255.250
PING 10.47.255.250 (10.47.255.250): 56 data bytes
64 bytes from 10.47.255.250: seq=0 ttl=63 time=1.750 ms
64 bytes from 10.47.255.250: seq=1 ttl=63 time=0.815 ms
64 bytes from 10.47.255.250: seq=2 ttl=63 time=0.851 ms
64 bytes from 10.47.255.250: seq=3 ttl=63 time=0.809 ms
(snip)
64 bytes from 10.47.255.250: seq=34 ttl=63 time=0.853 ms
64 bytes from 10.47.255.250: seq=35 ttl=63 time=0.848 ms
64 bytes from 10.47.255.250: seq=36 ttl=63 time=0.833 ms
64 bytes from 10.47.255.250: seq=37 ttl=63 time=0.832 ms
64 bytes from 10.47.255.250: seq=38 ttl=63 time=0.910 ms
64 bytes from 10.47.255.250: seq=42 ttl=63 time=2.071 ms
64 bytes from 10.47.255.250: seq=43 ttl=63 time=0.826 ms
64 bytes from 10.47.255.250: seq=44 ttl=63 time=0.853 ms
64 bytes from 10.47.255.250: seq=45 ttl=63 time=0.851 ms
64 bytes from 10.47.255.250: seq=46 ttl=63 time=0.853 ms
64 bytes from 10.47.255.250: seq=47 ttl=63 time=0.851 ms
64 bytes from 10.47.255.250: seq=48 ttl=63 time=0.855 ms
64 bytes from 10.47.255.250: seq=49 ttl=63 time=0.869 ms
64 bytes from 10.47.255.250: seq=50 ttl=63 time=0.833 ms
64 bytes from 10.47.255.250: seq=51 ttl=63 time=0.859 ms
64 bytes from 10.47.255.250: seq=52 ttl=63 time=0.866 ms
64 bytes from 10.47.255.250: seq=53 ttl=63 time=0.840 ms
64 bytes from 10.47.255.250: seq=54 ttl=63 time=0.841 ms
64 bytes from 10.47.255.250: seq=55 ttl=63 time=0.854 ms
^C
--- 10.47.255.250 ping statistics ---
56 packets transmitted, 53 packets received, 5% packet loss
round-trip min/avg/max = 0.799/0.888/2.071 ms
/ #
-> 3 packet loss is seen (seq 38-42)
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.19.25 ctr nei
Introspect Host: 172.31.19.25
+----------------------+--------------+----------+----------+-----------+-------------+------------+------------+-----------+
| peer | peer_address | peer_asn | encoding | peer_type | state | send_state | flap_count | flap_time |
+----------------------+--------------+----------+----------+-----------+-------------+------------+------------+-----------+
| ip-172-31-13-9.local | 172.31.13.9 | 64512 | BGP | internal | Established | in sync | 0 | n/a |
+----------------------+--------------+----------+----------+-----------+-------------+------------+------------+-----------+
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.13.9 ctr nei
Introspect Host: 172.31.13.9
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
| peer | peer_address | peer_asn | encoding | peer_type | state | send_state | flap_count | flap_time |
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
| ip-172-31-19-25.local | 172.31.19.25 | 64512 | BGP | internal | Established | in sync | 0 | n/a |
| ip-172-31-25-102.local | 172.31.25.102 | 0 | XMPP | internal | Established | in sync | 0 | n/a |
| ip-172-31-33-175.local | 172.31.33.175 | 0 | XMPP | internal | Established | in sync | 0 | n/a |
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
[root@ip-172-31-13-9 ~]#
-> 新的控制器具有两个XMPP连接。
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.19.25 ctr route summary
检查主机：172.31.19.25
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
| name | prefixes | paths | primary_paths | secondary_paths | infeasible_paths |
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
| default-domain:default- | 0 | 0 | 0 | 0 | 0 |
| project:__link_local__:__link_local__.inet.0 | | | | | |
| default-domain:default-project:dci- | 0 | 0 | 0 | 0 | 0 |
| network:__default__.inet.0 | | | | | |
| default-domain:default-project:dci-network:dci- | 0 | 0 | 0 | 0 | 0 |
| network.inet.0 | | | | | |
| default-domain:default-project:default-virtual- | 0 | 0 | 0 | 0 | 0 |
| network:default-virtual-network.inet.0 | | | | | |
| inet.0 | 0 | 0 | 0 | 0 | 0 |
| default-domain:default-project:ip-fabric:ip- | 0 | 0 | 0 | 0 | 0 |
| fabric.inet.0 | | | | | |
| default-domain:k8s-default:k8s-default-pod-network | 0 | 0 | 0 | 0 | 0 |
| :k8s-default-pod-network.inet.0 | | | | | |
| default-domain:k8s-default:k8s-default-service- | 0 | 0 | 0 | 0 | 0 |
| network:k8s-default-service-network.inet.0 | | | | | |
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.13.9 ctr route summary
检查主机：172.31.13.9
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
| name | prefixes | paths | primary_paths | secondary_paths | infeasible_paths |
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
| default-domain:default- | 0 | 0 | 0 | 0 | 0 |
| project:__link_local__:__link_local__.inet.0 | | | | | |
| default-domain:default-project:dci- | 0 | 0 | 0 | 0 | 0 |
| network:__default__.inet.0 | | | | | |
| default-domain:default-project:dci-network:dci- | 0 | 0 | 0 | 0 | 0 |
| network.inet.0 | | | | | |
| default-domain:default-project:default-virtual- | 0 | 0 | 0 | 0 | 0 |
| network:default-virtual-network.inet.0 | | | | | |
| inet.0 | 0 | 0 | 0 | 0 | 0 |
| default-domain:default-project:ip-fabric:ip- | 7 | 7 | 2 | 5 | 0 |
| fabric.inet.0 | | | | | |
| default-domain:k8s-default:k8s-default-pod-network | 7 | 7 | 4 | 3 | 0 |
| :k8s-default-pod-network.inet.0 | | | | | |
| default-domain:k8s-default:k8s-default-service- | 7 | 7 | 1 | 6 | 0 |
| network:k8s-default-service-network.inet.0 | | | | | |
+----------------------------------------------------+----------+-------+---------------+-----------------+------------------+
[root@ip-172-31-13-9 ~]#
-> 老的控制器不再具有前缀。
在ISSU过程结束后，新的kube-manager 启动：
[root@ip-172-31-19-25 ~]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE
cirros-deployment-75c98888b9-6qmcm 1/1 Running 0 34m 10.47.255.249 ip-172-31-25-102.ap-northeast-1.compute.internal <none>
cirros-deployment-75c98888b9-lxq4k 1/1 Running 0 34m 10.47.255.250 ip-172-31-33-175.ap-northeast-1.compute.internal <none>
cirros-deployment2-648b98685f-b8pxw 1/1 Running 0 15s 10.47.255.247 ip-172-31-25-102.ap-northeast-1.compute.internal <none>
cirros-deployment2-648b98685f-nv7z9 1/1 Running 0 15s 10.47.255.248 ip-172-31-33-175.ap-northeast-1.compute.internal <none>
[root@ip-172-31-19-25 ~]#
-> 通过新的IP创建容器（10.47.255.247, 10.47.255.248 是来自新控制器的新地址）
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.19.25 ctr nei
Introspect Host: 172.31.19.25
+----------------------+--------------+----------+----------+-----------+--------+-----------------+------------+-----------------------------+
| peer | peer_address | peer_asn | encoding | peer_type | state | send_state | flap_count | flap_time |
+----------------------+--------------+----------+----------+-----------+--------+-----------------+------------+-----------------------------+
| ip-172-31-13-9.local | 172.31.13.9 | 64512 | BGP | internal | Active | not advertising | 1 | 2019-Jun-23 05:37:02.614003 |
+----------------------+--------------+----------+----------+-----------+--------+-----------------+------------+-----------------------------+
[root@ip-172-31-13-9 ~]# ./contrail-introspect-cli/ist.py --host 172.31.13.9 ctr nei
Introspect Host: 172.31.13.9
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
| peer | peer_address | peer_asn | encoding | peer_type | state | send_state | flap_count | flap_time |
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
| ip-172-31-25-102.local | 172.31.25.102 | 0 | XMPP | internal | Established | in sync | 0 | n/a |
| ip-172-31-33-175.local | 172.31.33.175 | 0 | XMPP | internal | Established | in sync | 0 | n/a |
+------------------------+---------------+----------+----------+-----------+-------------+------------+------------+-----------+
[root@ip-172-31-13-9 ~]#
-> 新控制器不再有iBGP路由到旧的控制器。旧控制器依然具有iBGP路由条目，虽然这个过程很快就要停止:)
在停止了就的控制器后，配置：
[root@ip-172-31-19-25 ~]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE
cirros-deployment-75c98888b9-6qmcm 1/1 Running 0 48m 10.47.255.249 ip-172-31-25-102.ap-northeast-1.compute.internal <none>
cirros-deployment-75c98888b9-lxq4k 1/1 Running 0 48m 10.47.255.250 ip-172-31-33-175.ap-northeast-1.compute.internal <none>
cirros-deployment2-648b98685f-b8pxw 1/1 Running 0 13m 10.47.255.247 ip-172-31-25-102.ap-northeast-1.compute.internal <none>
cirros-deployment2-648b98685f-nv7z9 1/1 Running 0 13m 10.47.255.248 ip-172-31-33-175.ap-northeast-1.compute.internal <none>
cirros-deployment3-68fb484676-ct9q9 1/1 Running 0 18s 10.47.255.245 ip-172-31-25-102.ap-northeast-1.compute.internal <none>
cirros-deployment3-68fb484676-mxbzq 1/1 Running 0 18s 10.47.255.246 ip-172-31-33-175.ap-northeast-1.compute.internal <none>
[root@ip-172-31-19-25 ~]#
-> 新容器依然可以被创建
[root@ip-172-31-25-102 ~]# contrail-status
Pod Service Original Name State Id Status
vrouter agent contrail-vrouter-agent running 9a46a1a721a7 Up 33 minutes
vrouter nodemgr contrail-nodemgr running 11fb0a7bc86d Up 33 minutes
vrouter kernel module is PRESENT
== Contrail vrouter ==
nodemgr: active
agent: active
[root@ip-172-31-25-102 ~]#
-> 具有新控制器的vRouter工作良好
/ # ping 10.47.255.250
PING 10.47.255.250 (10.47.255.250): 56 data bytes
64 bytes from 10.47.255.250: seq=0 ttl=63 time=1.781 ms
64 bytes from 10.47.255.250: seq=1 ttl=63 time=0.857 ms
^C
--- 10.47.255.250 ping statistics ---
2 packets transmitted, 2 packets received, 0% packet loss
round-trip min/avg/max = 0.857/1.319/1.781 ms
/ #
-> 在vRouter之间Ping成功了向后兼容
由于有好几种更新集群的方法（就地、ISSU、是否为ifdown vhost0），因此方法的选择也是个重要的话题。
在讨论细节之前，让我先描述一下vrouter-agent up / down的行为，以及ifup vhost0 / ifdown vhost0的行为。
重新启动vrouter-agent时，一种假设是重新创建了vrouter-agent容器和vhost0。
实际上，事实并非如此，因为vhost0与vrouter.ko是紧密耦合的，需要从kernel中卸载vrouter.ko的同时将其删除。所以从操作角度来说，需要ifdown vhost0，那么不仅需要更新vrouter-agent，还需要更新vrouter.ko。（ifdown vhost0也将在内部执行rmmod vrouter）。
因此，要讨论向后兼容，需要研究下面三个主题。
- controller与vrouter-agent的兼容性
- 如果没有向后兼容性，则需要ISSU
- vrouter-agent与vrouter.ko的兼容性
- 如果没有向后兼容性，则需要ifdown vhost0，这将导致最少5-10秒的流量损失，因此实际上意味着需要将流量转移到其它节点，如实时迁移（live migration）
- 由于vrouter-agent使用netlink与vrouter.ko同步数据，因此架构更改可能导致vrouter-agent发生意外行为（例如Ksync logic上的vrouter-agent分段错误）
- vrouter.ko与kernel的兼容性
-
如果没有向后兼容性，则需要更新kernel，因此这意味着需要将流量移至其它节点
- 当vrouter.ko具有不同的in-kernal API时，无法由kernel加载，并且无法创建vhost0和vrouter-agent
对于2和3，基于种种原因不可避免地需要进行kernel更新，因此一个可行的计划是，首先选择一个新的kernel版本，然后选择一个支持该kernel的vrouter-agent / vrouter.ko，再并检查当前使用的vrouter-agent是否可以与该版本的control一起使用。
- 如果运行良好，请使用就地更新；如果由于某些原因无法运行，或者需要回滚操作，此时需要使用ISSU
对于1，由于在导入config-api定义时ifmap会为每个版本维护white_list。
- voidIFMapGraphWalker::AddNodesToWhitelist(): https://github.com/Juniper/contrail-controller/blob/master/src/ifmap/ifmap_graph_walker.cc#L349
根据我的尝试，它似乎具有不错的向后兼容性（由于路由信息更新与BGP类似，因此大多数情况下也应该可以正常工作）。
为了验证这一点，我尝试使用不同版本的模块进行该设置，看起来仍然可以正常工作。
I-1. config 2002-latest, control 2002-latest, vrouter 5.0-latest, openstack queens
I-2. config 2002-latest, control 5.0-latest, vrouter 5.0-latest, openstack queens
II-1. config 2002-latest, control 2002-latest, vrouter r5.1, kubernetes 1.12
注意：不幸的是，这种组合不能很好地工作（cni无法从vrouter-agent获取端口信息），我想这是由于5.0.x和5.1之间的cni版本更改（0.2.0-> 0.3.1）引起的。
II-2. config 2002-latest, control 2002-latest, vrouter 5.0-latest, kubernetes 1.12
因此，即使不需要立即更改kernel和vRouter版本，比较频繁地更新config / control也是一个好习惯，以修复可能的错误。
Tungsten Fabric入门宝典系列文章——
Tungsten Fabric 架构解析系列文章——
第一篇：TF主要特点和用例
第二篇：TF怎么运作
第三篇：详解vRouter体系结构
第四篇：TF的服务链
第五篇：vRouter的部署选项
第六篇：TF如何收集、分析、部署？
第七篇：TF如何编排
第八篇：TF支持API一览
第九篇：TF如何连接到物理网络
第十篇：TF基于应用程序的安全策略
以上是关于Tungsten Fabric入门宝典丨关于集群更新的那些事的主要内容，如果未能解决你的问题，请参考以下文章
Tungsten Fabric入门宝典丨关于多集群和多数据中心
Tungsten Fabric入门宝典丨关于安装的那些事（上）
Tungsten Fabric入门宝典丨关于服务链BGPaaS及其它
Tungsten Fabric入门宝典丨关于服务链BGPaaS及其它