MasterofProject

Knowledge of Datacenter Atlassian products

53621 people read comment(2) Collection Report
Classification:

Knowledge of Datacenter Atlassian products

CloudOriginally is the cluster architecture, atlassian products, application development team to a wide fan and for many years, but the application as nodes (such as JIRA, confluence, bamboo... Applications such as) then cluster the operation team is less and less, hope to the point of view of the infrastructure cut, cited by following the breakdown of the guide you understand Altassian datacenter of the clustering scheme, and inventory to build datacenter, really need any third party technology to complete this architecture:



Clustering: an integral virtual environment (Virtualization):

To a node as a unit, virtual node of course than the entity nodes scheduling and cluster of the node group to migrate between entities, there is no ready virtual environment to how easily the completion of this task. To play the benefits brought by DataCeneter, the virtual environment is one of the necessary conditions, and does not need to run in a specific virtual products.


The virtual environment constructed by the famous virtual products can be

Write the picture here.


Double hot standby (Avaliable Hight)

In view of the load balance, perhaps the customer will enumerate the operation of the JIRA load is good, do not need to use the load balance. But the hot standby HA, not only in the performance or backup considerations, but is used. How do you say that? Believe in applications or plug-ins to upgrade the needs of everyone there, but in which node test it? In order to test to buy a set of applications? Even if the budget is abundant, the test is completed qualification may be personnel to meet the needs of the development, the next step is to how does not interfere in the online of the transferred? Or on the online direct update. Direct online update is most of the current practice, produced by the status of mistakes, sometimes also respond to not the original state, light is interrupted application developers, while missing data.
As a result, the datacenter applications plus pacemaker the establishment perfect dual hot preparation, so the application cluster or plug-in application will have a smooth evolution environment, related to the upgrade can let go of, front-end development team of demand can rapid response.


(pacemaker, corosync) dual hot standby Technology

Build up a brand new nodes, here recommend using CentOS or RedHat 7 is the Linux kernel is 3.10. In the implementation process has slightly compared to the previous RedHat or CentOS 5 ~ 6 is dependent on cli (command line) but the establishment perfect, no matter is between nodes and cut the speed change, or is some fault analysis are better able to stand the test.
Of course, to maintain the original HA Linux old environment, or the operating system version of the need to allow other software to rely on considerations and can not be upgraded, DataCenter architecture is still to be completed. Only in a multi node deployment, the best operating system chooses Linux as core to 3.10.

The following two are the pacemaker stack and the internal structure diagram:

Write the picture here.

In Pacemaker under the full construction of the hot standby HA, in the following three situations are normal operation of the.

Write the picture here.


Network file system (NFS) and storage device (DRBD)

Datacenter as a key Home-shared, originally atlassian series of applications have a local home, through the NFS the network file system for AP with a multiple nodes, to enjoy for a home directory. So collaborative operation, you can make a single node settings (such as plug-ins to upgrade or install), the other members of the cluster will follow the completion.

NFS supply end can be any cloud storage distributed file system technology, here with CEPH for example to explain, the front end is by NFS shared network file system, the back end is produced by three OSD nodes combination of raw device, next is an interesting combination, I put the monitor mon and metadata MSD original two nodes for a node in a total of two groups, so that it can be the creation of a ha NFS-server also is to provide a virtual IP to atlassian AP as a mount, so that your home will have a hot standby! Here the establishment of Pacemake HA is to NFS-server The service node mon and MSD combination, there exists the original mechanism of hot backup, do not need to rely on Packmaker HA.

Please refer to the following pattern to imagine my description of the arrangement:

Write the picture here.

Hot standby start we are only considering the application side, now let's look at how the database with dual hot manufacturing integrated, starts a course that the storage location for the database, integration of DRBD with pacemaker very tight, to it as the storage location for the database is ideal, however, the device between two nodes on a block level synchronization at any time with, but as a party will be mounted, so that architecture scheme is not suitable for load balancing for the database, only an appropriate database hot standby, ha.

DRBD provides only to raw device, again at the top of the file system is the operating system took over, next mount by pacemaker arrangement, DRBD module responsible for three visits (Driver Raw,Driver Disk,Driver NIC).

Write the picture here.

Write the picture here.

DRBDIs a very busy module,Because it is a block hierarchy., so although it is responsible for many of the face, but the efficiency is not reduced (visit the official website of DRBD, please connectHttp://drbd.linbit.com/
Remind again:DRBD storage structure, the cluster is only suitable for the integration of the hot standby architecture is HA,Not suitable load balancing arrangement, this point please note.


Load balance (LoadBalance)

Haproxy is a very good load balancing helmsman. In this role can with simple parameters to allocate, can make out in the face of a web-service required by the client base is to use what kind of model to operate. For example, roundrobin or static-rr, can even consider separately the underlying database is using leastconn mode.
Of course HAproxy is such an important gateway how can only by a single node as a new group, HA? A very interesting arrangement, the haproxy to group matches inside, whether by cloud storage group (SQL) or cloud computing group (HTML) can be, as long as the group in each member haproxy service is started, the deployment of parameters on pacemaker by the supply of virtual pseudo IP to face the demand of the client access, so haproxy have hot backup mechanism.

Can refer to the following picture to imagine this interesting arrangement.

Write the picture here.

  1. Roundrobin:Average polling in each node, can adjust the weights to different levels of server allocation of different load.
  2. Static-rr:The flat transmission is on each node, but the weight adjustment is invalid. No matter how each server specifications, according to the total number of nodes average polling.
  3. Leastconn:For longer sessions server, such as LDAP, SQL... And so on, after the adjustment will be slow to come into effect.
  4. First:The goal of the algorithm is to always use the minimum number of servers.
  5. Source:This ensures that the same client address will always reach the same server side.
  6. Url:With source difference in using URL to identify the client.
  7. Url_arm:With URL differences in the parameters, the general server access to the server.
  8. Hdr:This generally refers to is recognition of user header, also is the same a user account login to different client access server and are serviced by the same server.
  9. Rdp-cookie:Like HDR but cookie presence or not to determine whether to change the access to the host.


Setting steps (Setup DataCenter)

The following is the Atlassian original for JIRA DataCenter setup steps
Write the picture here.

  1. First, the database with the application (here refers to the JIRA) separation.
  2. Confirm that the original version of JIRA is more than 6.3 (inclusive), if the conditions are not satisfied, please upgrade or revision plan.
  3. Loadblance node will be arranged well, this article suggests that the HAproxy server.
  4. Share Home Directory Local, as one of the most important Share Home DataCenter location.
  5. Add second application nodes, the database settings, NFS-client set a good access to Home Shared.
  6. The upper haproxy AP set to compose mode, the underlying database is set lessconn, then all connected in series, by pacemaker build good virtual IP, ready to accept client access.


Performance monitoring tool (Monitor Performance)

In the whole process of the implementation DataCentor fairly cumbersome, not only many steps and framework of diversification, but the model is consistent, will construct datacenter JIRA application also will be to construct the datacenter of other applications. Throughout the process, here advice at any time observation of virtual environment performance monitoring tools, while build while diagnosis and don't most of the construction steps to check back problems, it will increased problem analysis of the degree of difficulty.

The following icon example, in our view of jiraIP we can know that HA1 and HA2 are not roundrobin relationship between each other
Write the picture here.

In after the completion of the whole datacenter to build on the start line, operation process should pay attention to the CPU negative do not wear continues to rise in more than 80%.The rising need tuning or hardware upgradesThe moment is not in the way.

As shown in the VMware performance monitor only an instant rise.
Write the picture here.


In this paper the author: XU James (CSDN senior technical advisor) without permission shall not be reproduced!
Software mall:Http://prog3.com/sbdm/mall/tool
Hot activity:Atlassian purchase software technology Services Limited special offers hot!

top
Zero
step on
Zero
Guess you're looking for
View comments
* the above user comments only represent their personal views, does not represent the views or position of the CSDN website
    personal data
    • Visit160470 times
    • Integral:One thousand four hundred and seventy-five
    • Grade
    • Rank:16643rd name
    • Original14
    • Reproduced:33
    • Translation:5
    • Comments:45
    Latest comments