At AWS we provide services in availability zones, which are discrete
clusters of data centers strategically separated geographically within a
region, connected to other availability zones in the region by high-speed
high-bandwidth networks.
Across availability zones within a region, AWS offers RDS Oracle with
Multi-AZ, featuring sync writes across availability zones and automatic
failover. That gets you zero RPO within that setup and is a very popular
solution. RDS provides complete automation and management of provisioning,
backups, upgrades, performance monitoring and availability/DR.
So you might want to look into whether that level of geographic separation
with zero RPO works for you. If you wanted to add an additional
cross-region solution on top of that with less than zero RPO, you could do
that.
In general, zero RPO across regions, in or out of the cloud, is fraught. It
introduces often unacceptable latency, and any network partition event
results in either greater than zero RPO or complete loss of transaction
commit on the primary side.
Jeremiah
On Fri, Sep 23, 2022 at 9:50 AM Tim Gorman <tim.evdbt@xxxxxxxxx> wrote:
Chris,
Whether you're on-prem or in cloud...
- RAC, RAC one-node, and OS-level HA clusters (i.e.
Pacemaker/Corosync, etc) provide protection only to the *database *
*service*, not to the *data* in the database
- RPO isn't a factor with service protection, as RPO is a characteristic
of *data protection*
- Restore/recovery from database backups can not guarantee RPO lower than
the initialization parameter setting for ARCHIVE_LAG_TARGET
- ...and there are many reasons to consider RPO to be at least
twice the ARCHIVE_LAG_TARGET value as well...
- Data replication utilities like GoldenGate, SharePlex, etc cannot
guarantee RPO=0 due to all the queuing and forwarding necessary
- "*eventual consistency*" also implies that "*eventual*" could
turn out to be "*never*"
The only Oracle mechanism which guarantees RPO=0 is DataGuard using MAX
PROTECTION mode, while DataGuard in MAX AVAILABILITY mode comes in with a
close second.
Even the Zero Data Loss Recovery Appliance (ZDLRA) essentially employs the
same mechanisms used by DataGuard MAX PROTECTION mode.
Please let me know what you think?
Thanks!
-Tim
On 9/23/2022 3:51 AM, Chris Taylor wrote:
Hey guys, recently started working at a new company that's moving their
Oracle database to AWS.
One of the things they want is a near-0 RPO.
I was looking into AWS snapshot and replication to accomplish this which
seems pretty doable.
Either in the same region or into another region in case the region was
down/unavailable.
So I'm looking for two things:
1.) Other options? Active-Passive setup in 2 different regions? (Like
Failover but non-RAC)
2.) Anyone using AWS and snapshot copies & replication for Oracle in AWS?
Reference:
(Taking snapshots)
https://aws.amazon.com/blogs/database/improving-oracle-backup-and-recovery-performance-with-amazon-ebs-multi-volume-crash-consistent-snapshots/
(Replicating Snapshots)
https://aws.amazon.com/blogs/aws/ebs-snapshot-copy/
Thanks,
Chris
