
Highly available. And just as fast after the next update.
HIS, lab system (LIS), radiology (RIS, PACS), PDMS or staff scheduling: many clinical applications keep their data in Microsoft SQL Server. We make sure these databases survive a server failure. We prepare updates, patches and version changes so that any slowdown in response times shows up immediately and is documented. Always coordinated with your software vendor.
What happens when the server fails? And is everything still as fast after the update?
When hospitals and hospital groups bring us in.
The vendor announces a major update.
A new release, a new database schema, perhaps a higher compatibility level. You want to be able to show afterwards whether anything got worse.
It has been slower since the last update.
The cause is unclear. We show which queries have changed.
An application needs to become highly available.
After an IT outage or an audit, it has to survive a server failure. What the application and the vendor support has yet to be clarified.
Failover was never tested with the application.
The cluster is in place, but nobody has checked whether the application reconnects after a failover.
A version upgrade is due.
SQL Server 2016 has received no security updates since July 14, 2026; 2017 gets them until October 12, 2027. Versions and end of support →
High availability your software vendor supports.
Always On availability group, failover cluster instance or a well-protected single instance: the right choice follows from what the application tolerates, what the vendor supports and how long the department can manage without the system.
Requirements first, then technology
Together with your IT team we establish how long a system may be down and how much data loss is acceptable. That determines whether an availability group, a failover cluster instance or a tested restore is the right fit.
What the edition can do
Standard Edition offers a two-node failover cluster instance or a basic availability group with one database and two replicas. Full availability groups with several databases require Enterprise Edition. Failover cluster or availability group? →
Everything outside the database
An availability group replicates databases, but not logins, Agent jobs or linked servers. We keep them identical on every node; otherwise the nightly job that feeds an interface is missing after a failover.
Test failover with the application
In a maintenance window we fail over together with IT and the vendor and check whether the application reconnects through the listener or the cluster name. Afterwards everyone knows how long the interruption lasts.
Patching with a short interruption
With two nodes, the passive node is patched first, then you fail over. The application is down only for the failover, not for the whole installation.
See what an update changes.
A vendor update, a cumulative update or a higher compatibility level can change the execution plans of individual queries. Often nobody notices; sometimes an important query becomes noticeably slower. Query Store makes this visible before it turns into an ongoing problem.
- Baseline before
Query Store records in the application database which queries run, for how long and with which plan. If it is not running there yet, we enable it with the vendor's approval well before the update. - Comparison after
After the update we compare the same queries: what got slower, since when, and which plan changed? For this we use our own tool, PSG QX. - An interim fix you can undo
If an earlier plan is demonstrably better, it can be forced, with the vendor's approval. We document it so it is reviewed and removed again at the next update. - Findings for the vendor
The query, both plans and the metrics go to the vendor, so their development team can fix the cause.
SQL Server check: the basics first.
High availability and stable response times require a well-maintained instance. This script shows within seconds the version and patch level, the cores SQL Server uses under its license, the memory limit, the last consistency check and full backup, the Agent, and the statistics of the application database.
On the right is the result on a simulated lab system, before and after the usual changes: SQL Server 2025 Standard with a test database, not a customer system.
edition Standard, SQL Server 2025 CU9 cores 24 logical processors detected, 24 used under the license memory max server memory not set (2147483647 MB) checkdb never completed successfully (1900-01-01) backup no full backup in msdb agent Stopped statistics created at 50,000 rows, 400,000 modifications since after changes limit 3,072 MB, CHECKDB and backup dated, statistics at 450,000 rows
Show the script (T-SQL, read-only)
/* =====================================================================
PSG · SQL Server in hospitals: baseline check for one instance
As of November 16, 2026 · www.psg.de/en/sql-server-hospitals/
READ-ONLY. The script changes nothing, but it reads system views and msdb.
Still: run it on a test or staging system first and coordinate with
operations and the software vendor. Many vendors specify in their
support terms who may work on the instance.
All queries run with OPTION (MAXDOP 1) so the check does not take up
several cores itself on a busy system.
NUR LESEND. Erst auf einem Testsystem ausführen und mit Betrieb und
Softwarehersteller abstimmen, bevor es auf dem Produktivsystem läuft.
Part A: instance (any database)
Part B: statistics (run in the application's database)
===================================================================== */
/* A1 · Version, edition and patch level */
SELECT SERVERPROPERTY('ProductVersion') AS version,
SERVERPROPERTY('ProductLevel') AS level,
SERVERPROPERTY('ProductUpdateLevel') AS cu,
SERVERPROPERTY('Edition') AS edition
OPTION (MAXDOP 1);
/* A2 · Cores: how many does SQL Server use, given the edition?
At startup SQL Server writes to the error log how many logical
processors it detected and how many it uses under its license, e.g.
"... 24 total logical processors; using 24 logical processors based on
SQL Server licensing." If the second number is smaller than the first,
the edition leaves cores of the VM unused. The message is in the log
written since the last restart; if the log has been cycled since then,
look in the older logs (1, 2, ...). */
EXEC sys.xp_readerrorlog 0, 1, N'licensing';
/* A3 · Memory: is a limit set?
2147483647 MB means no upper limit: SQL Server competes with the
operating system and other services on the same VM. */
SELECT (SELECT CAST(value_in_use AS bigint) FROM sys.configurations
WHERE name = N'max server memory (MB)') AS max_server_memory_mb,
i.physical_memory_kb / 1024 AS vm_memory_mb,
i.committed_target_kb / 1024 AS sql_target_mb
FROM sys.dm_os_sys_info AS i
OPTION (MAXDOP 1);
/* A4 · Databases: compatibility level, last consistency check, last full backup
last_checkdb = 1900-01-01 means: never checked successfully. */
SELECT d.name,
d.compatibility_level,
d.recovery_model_desc,
CAST(DATABASEPROPERTYEX(d.name, 'LastGoodCheckDbTime') AS datetime) AS last_checkdb,
(SELECT MAX(b.backup_finish_date) FROM msdb.dbo.backupset AS b
WHERE b.database_name = d.name AND b.type = 'D') AS last_full_backup
FROM sys.databases AS d
WHERE d.database_id > 4
ORDER BY d.name
OPTION (MAXDOP 1);
/* A5 · Is SQL Server Agent running? Without the Agent, no maintenance jobs run. */
SELECT servicename, status_desc, startup_type_desc
FROM sys.dm_server_services
OPTION (MAXDOP 1);
/* B1 · In the application's database: statistics with many modifications
since the last update (the 20 most conspicuous). */
SELECT TOP (20)
OBJECT_SCHEMA_NAME(st.object_id) + N'.' + OBJECT_NAME(st.object_id) AS table_name,
st.name AS statistic,
sp.last_updated AS last_updated,
sp.rows AS rows_at_update,
sp.modification_counter AS modifications_since
FROM sys.stats AS st
CROSS APPLY sys.dm_db_stats_properties(st.object_id, st.stats_id) AS sp
WHERE OBJECTPROPERTY(st.object_id, 'IsUserTable') = 1
ORDER BY sp.modification_counter DESC
OPTION (MAXDOP 1);What we change, what the vendor changes.
The application belongs to the vendor. From the start we separate what can be changed in the database within the approved scope from what only the vendor can change in their software.
Within the approved scope
Instance, high availability, maintenance and patch level: what your IT team or we may change is agreed with you and the vendor in advance.
Findings the vendor can work with
Query, plan, metrics and the conditions under which the problem occurs. We also work for vendors of clinical software and know what their development teams need.
You know where you stand
If a cause can only be fixed in the vendor's code, we name the remaining risk and the next verifiable step. After the vendor update we measure again to confirm it worked.
Critical applications first, then technology.
- 01
Identify critical applications
Which systems may be down for how long, and where does patient care depend on them directly?
- 02
Health check
We examine the two or three most important instances in depth: basics, availability and response times. What we check →
- 03
Align with the vendor
What they support, what they approve and in which order we proceed.
- 04
Implement
In agreed maintenance windows, with a fallback path and notice to the affected departments.
- 05
Prove it
Failover tested, response times compared before and after the update, results documented.
Start with the health check →
A limited first step: documented findings for your most important instances and a prioritized list of actions.
Build high availability →
Plan and build availability groups or failover cluster instances and hand them over with tested failover.
Keep one critical application fast →
With PSG MX we continuously watch performance and changes of a defined core application, across vendor updates too.
Short answers.
Does our vendor support an availability group?
That is in the vendor's system requirements, or you can ask. Some applications only run on a single instance or a failover cluster instance; others cannot connect through a listener. We work this out with you and the vendor before planning anything.
What do clinical staff notice during a failover?
A short interruption: open connections are dropped and the application has to reconnect. How long that takes, and whether the application manages it on its own, is what the joint test in the maintenance window shows.
How do we know whether an update made performance worse?
From a Query Store baseline recorded before the update. It should cover several days of normal operation, so we plan it together with the update date. Without a baseline, all you have is how users feel.
Which clinical applications do you work on?
All that keep their data in SQL Server: hospital information system (HIS), laboratory information system (LIS), radiology information system (RIS) and PACS, PDMS in intensive care, staff scheduling, pharmacy and administration. On the HIS we always work together with the HIS vendor and within their requirements.
What if SQL Server 2016 is still in use?
SQL Server 2016 has received no security updates since July 14, 2026. First we find out which version the vendor of your application approves. Then we plan the migration with application testing, an agreed cutover and a fallback path.
Can you take over database administration for our hospital?
Yes, with the PSG DBA Pool and together with your team: maintenance, monitoring, patching and backup checks for your SQL Server instances, within an agreed scope and at agreed times.
Which application must not fail?
In the first conversation we work out which application it is, what the vendor supports and which step to take first.