VMware Cloud Community
ManivelR
Hot Shot
Hot Shot
Jump to solution

VCD HA(active and passive setup) with DRDB

Hi All,

VCD HA(active and passive setup)

I have some queries about vCloud director HA setup.Could you please help

on this?

We have setup vcloud director software(VCD 9.5.0) on 2
VMs(active/passive) with Centos 7.10 also installed Postgress on 2
VMs(active/passive) with Centos 7.10.

Setup is as follows:-

VCD application Setup
2 nodes, with CentOS-7.10, active/passive
/Opt filesystem for vCloud binary installation used, which is managed
by DRBD to replicate to other node

Clustering software used
Corosync, pacemaker

Using pacemaker,Virtual IP will be assigned for https service.

DB Setup:
1) 2 nodes, with CentOS-7.10, active/passive
2) /var/lib/pgsql filesystem for postgresql-9.5 database, which is
managed by DRBD to replicate to other node
3) Clustering software used :
Corosync, pacemaker & DRBD
Using pacemaker Virtual IP will be assigned for DB instance

Can we go with this setup or it should be routed through NFS mount
point only ? Please let me know which approach is correct one ?

Thank you,

Manivel R

1 Solution

Accepted Solutions
sk84
Expert
Expert
Jump to solution

Maybe it works, maybe it doesn't and you have replay or splitbrain problems. In any case, this setup is not as designed by the HA concept of vCD. vCD can easily be clustered with multiple active cells, which can also be load balanced, and for HA of the database you have to use the HA mechanisms of the DBMS (MSSQL, Postgres, etc.).

DRBD, Pacemaker and Corosync are basically not bad solutions, but they are not supported by any professional software. They are more a DIY solution.

Finally, you need to explain to your customers why the vCD environment is broken, configurations have been executed multiple times, and in the worst case data loss and application failures occur when both cells become active and the database is not designed for multi-master write operations.

From my experience with these products I can tell you the following:

- 2 pacemaker nodes are not enough, because they're vulnerable to splitbrain scenarios when both nodes become active. For quorum you should use a third node or a stonith device.

- vCD is a Java application that needs some time to start. Here you should set the timeout values of the pacemaker service accordingly high.

--- Regards, Sebastian VCP6.5-DCV // VCP7-CMA // vSAN 2017 Specialist Please mark this answer as 'helpful' or 'correct' if you think your question has been answered correctly.

View solution in original post

4 Replies
sk84
Expert
Expert
Jump to solution

Maybe it works, maybe it doesn't and you have replay or splitbrain problems. In any case, this setup is not as designed by the HA concept of vCD. vCD can easily be clustered with multiple active cells, which can also be load balanced, and for HA of the database you have to use the HA mechanisms of the DBMS (MSSQL, Postgres, etc.).

DRBD, Pacemaker and Corosync are basically not bad solutions, but they are not supported by any professional software. They are more a DIY solution.

Finally, you need to explain to your customers why the vCD environment is broken, configurations have been executed multiple times, and in the worst case data loss and application failures occur when both cells become active and the database is not designed for multi-master write operations.

From my experience with these products I can tell you the following:

- 2 pacemaker nodes are not enough, because they're vulnerable to splitbrain scenarios when both nodes become active. For quorum you should use a third node or a stonith device.

- vCD is a Java application that needs some time to start. Here you should set the timeout values of the pacemaker service accordingly high.

--- Regards, Sebastian VCP6.5-DCV // VCP7-CMA // vSAN 2017 Specialist Please mark this answer as 'helpful' or 'correct' if you think your question has been answered correctly.
ManivelR
Hot Shot
Hot Shot
Jump to solution

Thank you Sebestian for your valuable inputs.It is really helpful.

As communicated earlier( we have two cells for VCD and 2 VMs for Postgress DB) and so far we have not introduced any RABBITMQ AMQP.

When we use two VCD cells(Active/passive),these two cells can communicate directly to each other via AMQP or any other separate RABBITMQ AMQP virtual machine is required ?

Please advise.

Thank you,

Manivel RR

Reply
0 Kudos
sk84
Expert
Expert
Jump to solution

For the communication between the cells you need an AMQP service like RabbitMQ. See: vCD 9.5 (vCD 8.20) when we need RabbitMQ ? and when we can skip this component ?

However, if you only use one cell actively and the other cell is passive, this is not so important because all configuration and state information are stored in the database.

Only when multiple cells are used simultaneously is it important that the cells exchange certain events and task information, for example, when multiple people are doing configurations at the same objects.

--- Regards, Sebastian VCP6.5-DCV // VCP7-CMA // vSAN 2017 Specialist Please mark this answer as 'helpful' or 'correct' if you think your question has been answered correctly.
ManivelR
Hot Shot
Hot Shot
Jump to solution

Thank you very much Sebastian.

Reply
0 Kudos