docs: Direct DB Connection docs WIP

docs/sources/reference/direct-db-connection.md

@@ -0,0 +1,43 @@
+# Direct DB Connection
+
+Since version 4.3 Grafana has MySQL data source, Grafana-Zabbix plugin can use it for querying data directly from
+Zabbix database.
+
+One of the most hard queries for Zabbix API is history queries. For long time intervals `history.get`
+returns huge amount of data. In order to display it, plugin should adjust time series resolution
+by using [consolidateBy](/reference/functions/#consolidateby) function. Ultimately, Grafana displays this reduced
+time series, but that data should be loaded and processed on the client side first. Direct DB Connection solves this
+two problems by moving consolidation to the server side. Thus, client get ready-to-use dataset which has much smaller
+size. Data loads faster and client don't spend time for data processing.
+
+Also, many users point better performance of direct database queries versus API calls. This caused by several reasons,
+such as additional PHP layer and additional SQL queries (user permissions checks).
+
+## Query structure
+
+Grafana-Zabbix uses queries like this for getting history:
+
+```sql
+SELECT itemid AS metric, clock AS time_sec, {aggFunc}(value) as value
+  FROM {historyTable}
+  WHERE itemid IN ({itemids})
+    AND clock > {timeFrom} AND clock < {timeTill}
+  GROUP BY time_sec DIV {intervalSec}, metric
+```
+
+where `{aggFunc}` is one of `[AVG, MIN, MAX, SUM, COUNT]` aggregation function, `{historyTable}` is a history table,
+`{intervalSec}` - consolidation interval in seconds.
+
+When getting trends, plugin additionally queries particular value column (`value_avg`, `value_min` or `value_max`)
+depends on `consolidateBy` function value:
+
+```sql
+SELECT itemid AS metric, clock AS time_sec, {aggFunc}({valueColumn}) as value
+  FROM {trendsTable}
+  WHERE itemid IN ({itemids})
+    AND clock > {timeFrom} AND clock < {timeTill}
+  GROUP BY time_sec DIV {intervalSec}, metric
+```
+
+As you can see, plugin uses aggregation by given time interval. This interval is provided by Grafana and depends on the
+panel with in pixels. Thus, Grafana always gets data in necessary resolution.