Understanding OceanBase Clog Directory Structure and Disk Usage Control

In OceanBase the /data/log1/clustername/clog directory stores physical log files (clog) that record database modifications. Starting with version 4.1.0, OceanBase introduces tenant‑level log streams, grouping physical changes into a system log stream and multiple user log streams that are synchronized via the Paxos protocol.

The original clog files are no longer placed directly under the OBServer node; instead two new layers appear: log_pool and tenant_id directories. The log_pool pre‑allocates clog files based on the node's total log disk capacity ( LOG_DISK_CAPACITY ) and the default clog size of 64 MiB. Allocation follows the configuration items log_disk_size and log_disk_percentage :

If log_disk_size is 0 and log_disk_percentage is non‑zero, the percentage value determines the allocated space.

If log_disk_size is non‑zero, it overrides the percentage setting.

If both are zero, OceanBase automatically assigns 30 % of the disk to Redo logs when the log and data share the same disk, otherwise 90 % when they are on separate disks.

The system table __all_virtual_server records the total log‑disk capacity ( log_disk_capacity ) and two important fields: log_disk_assigned (pre‑allocated space in the log pool) and log_disk_in_use (actual space used by a tenant). These values are reflected in the view gv$ob_units for each tenant unit.

Each tenant appears as a tenant_id sub‑directory under clog . The space is not allocated to the tenant immediately; it is reserved in the log pool and becomes visible in log_disk_assigned . When a tenant writes new clog files, OceanBase moves the reserved files to the tenant’s directory, updating log_disk_in_use . The meta‑tenant (e.g., tenant_1001 ) is automatically created for every user tenant starting from version 4.0.0 to store internal metadata.

Below the tenant_id directory there is an additional numeric ID directory that corresponds to the log‑stream ID (LS_ID). The view gv$ob_log_stat can be used to query a tenant’s log‑stream IDs, while CDB_OBJECTS and CDB_OB_TABLET_TO_LS map tables to tablets and tablets to log streams, respectively.

Clog disk usage is limited per tenant. When the utilization ratio (log_disk_in_use / log_disk_size * 100%) reaches the default threshold of 80 % (non‑configurable), the system stops requesting new clog files from the log pool and reuses the oldest ones.

To monitor clog generation and disk consumption, the article provides a Bash script that counts clog files per tenant, calculates their total size, and aggregates hourly and daily statistics using find , stat , and awk . Sample output shows the remaining clog files in the log pool and per‑tenant usage.

References:

[1] LOG_DISK_CAPACITY documentation

[2] LOG_DISK_SIZE documentation

[3] Meta tenant concept

[4] Log stream documentation