Cluster migration advice needed

[benhank]

Hello all,

I have a project:
1. Migrate my NLS 2 node cluster to new equipment
a. the new cluster will have different hostnames and IP addresses then the current cluster.
2. Migrate the currently stored data to new equipment with minimal gaps in data collection
3. Place the new cluster behind a VIP.

Can I get some advice on how to make this happen?

[gsmith]

Hi,

Here's one idea:

1. Take an outage
2. Re-IP the servers in the old cluster, verify functionality
3. Create a VIP with the old Log Server master's IP, put the old cluster servers in a pool tied to the VIP
4. Bring the VIP online (outage over) - logging should continue as before
5. stand up new NLS master and slave
6. restore data from current master NLS to new NLS master and restore data from current slave NLS to new NLS slave
(step 6 could take days)
7. repeat step 6 until data on new NLS servers is close to the data on the current NLS servers
8. take an outage, do a final restore from current NLS servers to new NLS servers
9. add new NLS servers to pool behind VIP, remove old NLS servers from VIP
10. go back online

If your servers were using NFS mounts for data you could replace steps 6 thru 8
with:
6. take an outage
7. unmount data drives from current NLS servers
8. mount data drives to new NLS servers
9. add new NLS servers to pool behind VIP, remove old NLS servers from VIP
10. go back online

You need to practice this on a test system in order to get the full-detailed steps identified, and to
get some timings so you know how long you will be down. There are a lot of validations that need to
be part of the plan(s), this is just a high-level idea.

Let me know what you think.

Thanks

[benhank]

First of all: Man, This is awesome! THANK YOU!
I have a limitation,
The VIPs we use are pulled from a specific range of ip's. I can't use the ip of my NLS server.

Will this work?
1. place current cluster behind a vip (round robin)
2. point all sending devices to the vip, verify that the data is being collected.
3. pull 1 server from cluster and replace with new
4. wait for data to replicate
5. repeat

I get that NLS is load-balanced, but how does NLS handle a round robin setup?

if data is being sent to each server in the cluster does it cause a performance hit, or are there any gotchas that i should be aware of?
again thanks for the input so far!

[gsmith]

Hi,

Answers/comments:

Will this work? Yes

1. place current cluster behind a vip (round robin)

2. point all sending devices to the vip, verify that the data is being collected.
This will most likely take a bit of time.

3. pull 1 server from cluster and replace with new
I'd add two new servers, wait for data to get re-distributed.

4. wait for data to replicate
Remove one old server, wait for data to get re-distributed.
Remove second old server - preferably over the weekend.


I get that NLS is load-balanced, but how does NLS handle a round robin setup?
For incoming data the round robin should be fine. When accessing the web interface it's probably best
to go to one of the two nodes directly.

One of the gotcha's I can think of is that you'll need to make sure you can identify the true source of the
log files, otherwise they will all look like they are coming from the VIP

I strongly urge you run through this a few times with a Test set-up.

Thanks

[benhank]

One of the gotcha's I can think of is that you'll need to make sure you can identify the true source of the
log files, otherwise they will all look like they are coming from the VIP

dang..

[gsmith]

What are you using for a VIP? If I remember correctly with
the F5 you could create create rewrite rules to include the
source IP.

Thanks