Products / yuzuy software / Failover / Failback

yuzuy software · Failover / Failback

Failover for Qumulo in one checked run, and back.

yuzuy switches all replication relationships of a cluster pair to the other side together, with checks at every step. Failback brings them home the same way. Qumulo replicates the data; yuzuy orchestrates everything around it.

Status

How it works

The same steps, in the same order, every time.

Failover tends to happen when things are hectic. yuzuy knows every step and runs them for you, instead of a checklist of CLI commands and API calls.

  1. Running replication jobs finish where possible, and yuzuy saves the replication state.
  2. Source directories on the primary become read-only, and their shares are renamed.
  3. The selected floating IPs move to the secondary; networks are created there if needed.
  4. Replicated directories on the secondary become writable. Clients carry on.

Checks run throughout, and the web interface shows the status of every step.

Fact sheet

StartOne click in the web interface, or one API call, for example from your monitoring
ScopeAll replication relationships of a relation switch together
DurationTypically about 5 seconds; running transfers get a configurable timeout (default 1 minute)
NetworkIP failover moves floating IPs to the secondary (mostly NFS); DFS failover removes them on the primary (SMB via DFS)
FailbackKeep data (default), replicate changed data back, or discard changes
ClustersAny number of clusters and relations; cross-replicating pairs supported
RecordProgress and history per relation
CostIncluded in the yuzuy subscription

Changelog

See it in action

A planned and an unplanned failover, step by step.

Click play to load the video from YouTube (Google). Only then does YouTube receive data such as your IP address and may set cookies. Privacy policy · Watch on YouTube
Click play to load the video from YouTube (Google). Only then does YouTube receive data such as your IP address and may set cookies. Privacy policy · Watch on YouTube

All videos

No half measures

No test mode, on purpose.

yuzuy has no dry run and no partial failover. A test that skips steps proves little and gives a false sense of safety. A planned failover is the real test, and yuzuy makes it a routine one.

Prepare once

What to set up outside yuzuy.

  • Separate FQDNs for NFS, SMB without DFS and SMB via DFS
  • A plan for distributing floating IPs (older Qumulo Core versions limit them per node and VLAN)
  • A DFS cache duration that suits your failover
  • Target directories named after their source cluster
  • Admin rights on both clusters

“The collaboration between Qumulo and yuzuy has not only met but exceeded the hospital’s expectations.”

Alex Weber
Storage Systems Engineer, Platform Services
Universitätsspital Basel

Universitätsspital Basel has run failover and failback for its Qumulo clusters with yuzuy since 2023.

Read the reference

Background

Read more on the topic.

Article

How to fail over Qumulo Continuous Replication for SMB and NFS clients

The steps of a planned and an unplanned failover, the failback, and where failovers go wrong.

Read the article

Article

Qumulo failover with DFS namespaces

Why DFS fails over on its own but rarely fails back, and how to bring SMB clients home reliably.

Read the article

Article

Keeping shares, exports and quotas in sync between replicated clusters

Replication copies data, not configuration: by hand, by script, or automatically.

Read the article

FAQ

Questions admins ask.

Are there any limitations to the number of floating IP addresses that Qumulo or yuzuy have?

Qumulo limits the number of floating IP addresses per storage node per VLAN to 10. This applies especially in failover scenarios where a cluster with many nodes fails over to one with fewer nodes, as typically all floating IPs from the primary cluster must be transferred to the secondary.

In failover relationships with significant differences in the number of nodes between clusters, there’s a risk of exceeding the 10 floating IP limit on the secondary cluster, leading to errors. yuzuy checks for this during configuration and alerts you. Proper planning of floating IP distribution from the start is recommended to avoid DNS changes later. yuzuy also assists with a planning tool for IP address distribution.

Link to this answer
How long does a failback take?

During a failback, you can choose to replicate new data back to the primary cluster or discard it. Typically, new data is replicated back, with the duration depending on the amount of data and performance. Clients remain connected to the secondary cluster without interruption. Once the primary cluster is up to date and in continuous replication state, the final step of the failback takes only a few seconds.

Link to this answer
What happens to replication data in flight when initiating a failover?

For planned failovers and failbacks, yuzuy waits for replication to complete before proceeding. The timeout for this can be configured, and if necessary, yuzuy can cancel the ongoing replication and force a failover. In the event of a cluster outage, some data in flight may not have been replicated yet.

Link to this answer

All questions on replication and failover

Talk to the people who build it.

Questions, a demo or a quote: we reply within one business day.