Sign in
Server health

See the host, not only what the database login sees

Every connection gets a health card: CPU, memory, disk, queries per second and a 48-hour trend. A database login cannot see its machine, so Migratrix can read the host's own figures from the monitoring you already run: Prometheus, Zabbix or Datadog, or CloudWatch, Azure Monitor and Google Cloud Monitoring for managed databases.

What a database login can tell you about its host

Queries per second, sessions and cluster instances always come from the database. A metrics source fills in what the login cannot see; a figure it does not have stays the database's own.

Engine From the login alone With a metrics source
PostgreSQL No CPU, memory or disk figure. The card shows active sessions. Host CPU, memory, disk, disk IOPS and network.
MySQL, MariaDB, TiDB No CPU. Memory is the InnoDB buffer pool, disk the tablespace files. Host CPU, memory and disk in their place.
Oracle Host CPU per instance, including RAC. Memory is SGA and PGA, disk the data files or ASM disk groups. The machine's memory and file systems.
SQL Server CPU, host memory and the volumes that hold database files. The whole machine's figures, and network.

Six sources, read from inside your network

Prometheus
Any Prometheus-compatible API: Prometheus, Thanos, Grafana Mimir, VictoriaMetrics, and the managed services of AWS, Azure and Google Cloud, signed with that cloud's own identity. The defaults read node_exporter; your own PromQL replaces them when your exporters or labels differ.
Managed databases
CloudWatch for RDS and Aurora, Azure Monitor for Azure SQL Database, Managed Instance and the PostgreSQL and MySQL flexible servers, Google Cloud Monitoring for Cloud SQL. The instance is found from the connection's own address, with the Executor's cloud identity or a key you give it, read-only.
Zabbix and Datadog
The hosts they already monitor. Zabbix through its API, by the items of the standard agent templates, 5.0 and later, entirely inside a network with no internet access. Datadog through its query API, by the system metrics of the Datadog Agent.
Finding each server
By the connection's host name, with no setup, or by the name or label matcher your monitoring uses. A test button reads the figures before anything is saved.
On the card
The host's figures decide the health label and the Server Alerts count, with "via Zabbix", "via Azure Monitor" and so on so nobody guesses where a number came from. If the source does not answer, the card says so and falls back to the database's figures.
The Executor reads, the cloud does not

The Executor next to your databases queries the monitoring system. Its address only has to be reachable from there, and its password, token or AWS key is kept encrypted on the Executor host with your database credentials.

Installed like an engine

Each kind is a package on the Executor, verified, and signature-checked where you require it, by the same rules as an engine plugin. A package brings its own form, and the cloud SDKs live only in the packages that need them, not in the Executor itself.

Read-only and audited

Nothing is installed in your databases or written to your monitoring. Adding, changing or deleting a source is recorded in the audit ledger, with which secret changed and never its value.

What it is, and what it is not

In the product Not claimed
A health card per connection, sampled about once a minute, with a 48-hour hourly trend and filters by environment and health. A monitoring product. Keep your dashboards and on-call alerting; this is the view next to the change you are about to run.
Host figures from Prometheus-compatible APIs (including the managed services of AWS, Azure and Google Cloud), Zabbix, Datadog, CloudWatch, Azure Monitor and Google Cloud Monitoring. AlloyDB, Azure SQL elastic pools, New Relic or Dynatrace as sources.
One set of thresholds for the health label: 75% for Warning, 90% for Critical. Threshold notifications by email or Slack, or thresholds per server.

Point it at your own Prometheus

Bring the address of your monitoring. We will show your database hosts on the cards before the call ends.

All features