Secondary WAL failover store must be durable

src/current/v24.1/wal-failover.md

@@ -472,6 +472,12 @@ Store _A_ will failover to store _B_, store _B_ will failover to store _C_, and
 
 However, the WAL failback operation will not cascade back until **all drives are available** - that is, if store _A_'s disk unstalls while store _B_ is still stalled, store _C_ will not failback to store _A_ until _B_ also becomes available again. In other words, _C_ must failback to _B_, which must then failback to _A_.
 
+### 13. Can I use an ephemeral disk for the secondary storage device?
+
+No, the secondary (failover) disk **must be durable and retain its data across VM or instance restarts**. Using an ephemeral volume (for example, the root volume of a cloud VM that is recreated on reboot) risks permanent data loss: if CockroachDB has failed over recent WAL entries to that disk and the disk is subsequently wiped, the node will start up with an incomplete [Raft log]({% link {{ page.version.version }}/architecture/replication-layer.md %}#raft) and will refuse to join the cluster. In this scenario the node must be treated as lost and replaced.
+
+Always provision the failover disk with the same persistence guarantees as the primary store.
+
 ## Video demo: WAL failover
 
 For a demo of WAL Failover in CockroachDB and what happens when you enable or disable it, play the following video:

src/current/v24.3/wal-failover.md

@@ -470,6 +470,12 @@ Store _A_ will failover to store _B_, store _B_ will failover to store _C_, and
 
 However, the WAL failback operation will not cascade back until **all drives are available** - that is, if store _A_'s disk unstalls while store _B_ is still stalled, store _C_ will not failback to store _A_ until _B_ also becomes available again. In other words, _C_ must failback to _B_, which must then failback to _A_.
 
+### 13. Can I use an ephemeral disk for the secondary storage device?
+
+No, the secondary (failover) disk **must be durable and retain its data across VM or instance restarts**. Using an ephemeral volume (for example, the root volume of a cloud VM that is recreated on reboot) risks permanent data loss: if CockroachDB has failed over recent WAL entries to that disk and the disk is subsequently wiped, the node will start up with an incomplete [Raft log]({% link {{ page.version.version }}/architecture/replication-layer.md %}#raft) and will refuse to join the cluster. In this scenario the node must be treated as lost and replaced.
+
+Always provision the failover disk with the same persistence guarantees as the primary store.
+
 ## Video demo: WAL failover
 
 For a demo of WAL Failover in CockroachDB and what happens when you enable or disable it, play the following video:

src/current/v25.2/wal-failover.md

@@ -470,6 +470,12 @@ Store _A_ will failover to store _B_, store _B_ will failover to store _C_, and
 
 However, the WAL failback operation will not cascade back until **all drives are available** - that is, if store _A_'s disk unstalls while store _B_ is still stalled, store _C_ will not failback to store _A_ until _B_ also becomes available again. In other words, _C_ must failback to _B_, which must then failback to _A_.
 
+### 13. Can I use an ephemeral disk for the secondary storage device?
+
+No, the secondary (failover) disk **must be durable and retain its data across VM or instance restarts**. Using an ephemeral volume (for example, the root volume of a cloud VM that is recreated on reboot) risks permanent data loss: if CockroachDB has failed over recent WAL entries to that disk and the disk is subsequently wiped, the node will start up with an incomplete [Raft log]({% link {{ page.version.version }}/architecture/replication-layer.md %}#raft) and will refuse to join the cluster. In this scenario the node must be treated as lost and replaced.
+
+Always provision the failover disk with the same persistence guarantees as the primary store.
+
 ## Video demo: WAL failover
 
 For a demo of WAL Failover in CockroachDB and what happens when you enable or disable it, play the following video:

src/current/v25.3/wal-failover.md

@@ -470,6 +470,12 @@ Store _A_ will failover to store _B_, store _B_ will failover to store _C_, and
 
 However, the WAL failback operation will not cascade back until **all drives are available** - that is, if store _A_'s disk unstalls while store _B_ is still stalled, store _C_ will not failback to store _A_ until _B_ also becomes available again. In other words, _C_ must failback to _B_, which must then failback to _A_.
 
+### 13. Can I use an ephemeral disk for the secondary storage device?
+
+No, the secondary (failover) disk **must be durable and retain its data across VM or instance restarts**. Using an ephemeral volume (for example, the root volume of a cloud VM that is recreated on reboot) risks permanent data loss: if CockroachDB has failed over recent WAL entries to that disk and the disk is subsequently wiped, the node will start up with an incomplete [Raft log]({% link {{ page.version.version }}/architecture/replication-layer.md %}#raft) and will refuse to join the cluster. In this scenario the node must be treated as lost and replaced.
+
+Always provision the failover disk with the same persistence guarantees as the primary store.
+
 ## Video demo: WAL failover
 
 For a demo of WAL Failover in CockroachDB and what happens when you enable or disable it, play the following video: