@Rocky Are the rotation settings for the following log files applied globally based on the CN configuration? •

{be or cn}.INFO.log

•

{be or cn}.WARN.log

•

{be or cn}.ERROR.log

•

{be or cn}.FATAL.log

For example, if the rotation settings are configured to roll at 1 GB and retain up to 5 files, does that limit apply independently to each log level? In that case, would the theoretical maximum disk usage be 20 GB in total?

INFO (1 GB × 5) + WARN (1 GB × 5) + ERROR (1 GB × 5) + FATAL (1 GB × 5) = 20 GB

Yes, your understanding is correct. The rotation settings for StarRocks Backend (BE) and Compute Node (CN) logs apply independently to each log level stream based on the global configuration in

be.conf

(or

cn.conf

). 1. Independent Application Even though you define the settings once (e.g.,

sys_log_roll_num = 5

), the BE/CN process applies these limits to each log file type separately. StarRocks uses a version of

glog

where: *

<http://be.INFO|be.INFO>

maintains its own set of 5 rolled files. *

be.WARNING

maintains its own set of 5 rolled files. *

be.ERROR

maintains its own set of 5 rolled files. *

be.FATAL

maintains its own set of 5 rolled files. 2. Theoretical Maximum Disk Usage Your calculation is correct. If each of the four log levels reaches its rotation limit, the theoretical maximum disk usage for system logs would be:

1 GB (limit) × 5 (files) × 4 (levels: INFO, WARN, ERROR, FATAL) = 20 GB

Important Considerations: * Log Mirroring: In StarRocks (via

glog

), log levels are inclusive. An

ERROR

message is written to the

ERROR

,

WARN

, and

INFO

log files simultaneously. This means that during a period of high error volume, all three files will grow at once, consuming disk space for the same data across multiple files. *

be.out

Log: Note that there is also a

be.out

file which captures

stdout

and

stderr

. This file is not managed by the

sys_log_roll_num

or

sys_log_roll_mode

parameters. It typically only grows significantly if the process is crashing or printing non-standard output. * Import Error Logs: Logs related to data loading (stored in

storage/error_log

) are handled separately by the

load_error_log_reserve_hours

parameter and do not count toward this 20 GB total. Summary of Configuration Parameters: * `sys_log_roll_mode`: Set to

SIZE-MB-1024

for 1 GB rolls. * `sys_log_roll_num`: Set to

5

to retain 5 files per level. * `sys_log_dir`: The directory where these files are stored (default

${STARROCKS_HOME}/log

). References • administration/management/logs.md • administration/management/BE_configuration.md • administration/management/FE_configuration.md