added reference doc for cloud durability, closes #1915

content/influxdb/cloud/reference/internals/_index.md

@@ -0,0 +1,9 @@
+---
+title: InfluxDB Internals
+menu:
+  influxdb_cloud_ref:
+    name: InfluxDB Cloud internals
+weight: 7
+---
+
+{{< children >}}

content/influxdb/cloud/reference/internals/durability.md

@@ -0,0 +1,69 @@
+---
+title: InfluxDB Cloud data durability
+description: >
+  InfluxDB Cloud ensures the durability of all stored data by automatically creating
+  backups, replicating data across geographic locations, and verifying replicated
+  data is consistent and backups are readable.
+weight: 101
+menu:
+  influxdb_cloud_ref:
+    name: Data durability
+    parent: InfluxDB Cloud internals
+influxdb/cloud/tags: [backups, internals]
+---
+
+InfluxDB Cloud ensures the durability of all stored data by automatically creating
+backups, replicating data across geographic locations, and verifying replicated
+data is consistent and backups are readable.
+
+##### On this page
+
+- [Backup processes](#backup-processes)
+- [Recovery](#recovery)
+- [Data verification](#data-verification)
+
+## Backup processes
+InfluxDB Cloud stores **two out-of-band backup copies** of all data.
+The following processes each generate backups:
+
+- [Backup on write](#backup-on-write)
+- [Backup after compaction](#backup-after-compaction)
+
+### Backup on write
+All inbound write requests to InfluxDB Cloud queue in a durable message queue.
+The message queue stores the raw [line protocol](/influxdb/cloud/reference/glossary/#line-protocol)
+of each write request in an object storage location as an out-of-band backup.
+The message queue also writes data to the live storage service, replicates it twice
+across two geographic locations, and makes it available to the InfluxDB Cloud service.
+
+The durability of the message queue is 72 hours, meaning InfluxDB Cloud can sustain
+a failure of its underlying storage tier or object storage services for up to 72 hours.
+
+To minimize potential data loss due to defects introduced within the broader InfluxDB Cloud service,
+InfluxData minimizes code between the data ingest and backup processes.
+
+### Backup after compaction
+The InfluxDB storage engine compresses data over time in a process known as
+[compaction](/influxdb/cloud/reference/glossary/#compaction).
+When each compaction cycle completes, InfluxDB Cloud stores compressed
+[TSM](/influxdb/cloud/reference/glossary/#tsm-time-structured-merge-tree) files
+in an object storage location.
+
+## Recovery
+InfluxDB Cloud uses out-of-band backups stored in object storage to recover data:
+
+- **Message queue backup:** line protocol based on inbound write requests
+- **Historic backup:** compressed TSM files
+
+The Recovery Point Objective (RPO) is any accepted write.
+The Recovery Time Objective (RTO) is harder to definitively predict as potential failure modes can vary.
+While most common failure modes can be resolved within minutes or hours,
+catastrophic or extreme failure modes may take longer.
+For example, if we need to rebuild all data from the message queue backup,
+it could take 24 hours or longer.
+
+## Data verification
+InfluxDB Cloud has two data verification services running 24x7:
+
+- **Entropy detection:** ensures that replicated data is consistent
+- **Backup verification:** validates and ensures backups in object storage are readable

content/influxdb/v2.0/reference/glossary.md

@@ -188,6 +188,10 @@ Use comments with Flux statements to describe your functions.
 
 A standardized text file format used by the InfluxDB web server to create log entries when generating server log files.
 
+### compaction
+
+The process of optimizing disk usage by compressing time series data.
+
 ### continuous query (CQ)
 
 Continuous queries are the predecessor to tasks in InfluxDB 2.0.