Hello Team We have recently upgraded from 3 5 to 4 1 1 and n StarRocks #questions-and-troubleshooting

Hello Team, We have recently upgraded from 3.5 to ...

Mahesha Subrahamanya

07/31/2026, 1:47 AM

Hello Team, We have recently upgraded from 3.5 to 4.1.1 and noticed few changes mainly on the "DataCacheMetrics" i'e DiskUsage is 0B/0B however this used to work with 3.5 version. I'm trying to setup as we did in 3.5 version in 4.1.1 but still it does not show anything

Copy code

SHOW COMPUTE NODES;
Status: Normal, DiskUsage: 0B/0B, MemUsage: 94KB/2GB
Status: Normal, DiskUsage: 0B/0B, MemUsage: 97.2KB/2GB
Status: Normal, DiskUsage: 0B/0B, MemUsage: 763.5MB/2GB
Status: Normal, DiskUsage: 0B/0B, MemUsage: 67.8KB/2GB
Status: Normal, DiskUsage: 0B/0B, MemUsage: 1.9GB/2GB

Anybody have noticed this behavior , please let us know how to get this metrics back online or make it work meaning showing in the metrics. thanks. I tried to follow this change however i don't see any luck in getting this so - https://starrocks.slack.com/archives/C02FACZSNJV/p1785459758860569?thread_ts=1784643758.586899&cid=C02FACZSNJV

Mahesha Subrahamanya

07/31/2026, 4:07 AM

@Rocky any help on this question? thanks

Rocky

07/31/2026, 4:08 AM

Hi! I'm Rocky. I'd love to help, but it looks like the question didn't come through in your message. Could you please provide the details, such as an error message, a specific query, or the behavior you're seeing? Once you share that, I'll do my best to find the answer for you!

Mahesha Subrahamanya

07/31/2026, 4:08 AM

@Rocky We have recently upgraded from 3.5 to 4.1.1 and noticed few changes mainly on the "DataCacheMetrics" i'e DiskUsage is 0B/0B however this used to work with 3.5 version. I'm trying to setup as we did in 3.5 version in 4.1.1 but still it does not show anything

Copy code

SHOW COMPUTE NODES;
Status: Normal, DiskUsage: 0B/0B, MemUsage: 94KB/2GB
Status: Normal, DiskUsage: 0B/0B, MemUsage: 97.2KB/2GB
Status: Normal, DiskUsage: 0B/0B, MemUsage: 763.5MB/2GB
Status: Normal, DiskUsage: 0B/0B, MemUsage: 67.8KB/2GB
Status: Normal, DiskUsage: 0B/0B, MemUsage: 1.9GB/2GB

Rocky

07/31/2026, 4:09 AM

It appears you are encountering a behavior change introduced in StarRocks 4.0 regarding how Data Cache (formerly Block Cache) parameters are named and managed. In the transition from 3.5 to 4.1.1, several Data Cache configuration parameters were renamed to simplify the naming convention and unify the system. If you are using the old parameter names in your

be.conf

(or CN configuration), the cluster may be ignoring them, leading to a default disk quota of

0B

. 1. Key Parameter Renames in 4.0+ From StarRocks 4.0 onwards, several

datacache_

prefixed parameters for disk management were shortened. You should update your configuration to use the new names: Old Parameter (v3.x) | New Parameter (v4.0+)

datacache_auto_adjust_enable

| enable_datacache_disk_auto_adjust

datacache_disk_high_level

| disk_high_level

datacache_disk_safe_level

| disk_safe_level

datacache_disk_low_level

| disk_low_level
2. Check Automatic Scaling If

DiskUsage

is showing

0B/0B

, it usually means the system has not allocated any disk quota. • Auto-adjustment: If you want StarRocks to automatically manage disk space, ensure

enable_datacache_disk_auto_adjust

is set to

true

(this is the default in 4.1). • Manual Quota: If you want to set a fixed size as you did in 3.5, ensure

datacache_disk_size

is set (e.g.,

datacache_disk_size = 1T

). 3. Verify with
information_schema
To see exactly what the BE/CN thinks its configuration and paths are, run this query:

Copy code

sql
SELECT * FROM information_schema.be_datacache_metrics;

Check the

DISK_QUOTA_BYTES

and

DIR_SPACES

columns. If

DISK_QUOTA_BYTES

is 0, the configuration isn't being applied. If

DIR_SPACES

is empty, the BE cannot find or access the

storage_root_path

used for caching. 4. Troubleshooting Steps 1. Update Configs: Update your

be.conf

or CN config files to use the new parameter names mentioned above. 2. Verify Path: Ensure

storage_root_path

is correctly defined. For Compute Nodes (CN), this is mandatory for disk caching to work. 3. Check Disk Space: StarRocks v3.3+ (including 4.x) may skip enabling Data Cache automatically if the disk usage is already above the

disk_high_level

(default 90%) or if the remaining space is too low. 4. Dynamic Update: You can try to force a quota change at runtime to see if it responds:

Copy code

sql
   -- Example: Manually set disk size to 100GB for all nodes
   UPDATE be_configs SET VALUE="100G" WHERE NAME="datacache_disk_size";

If the issue persists, check the BE/CN leader logs (

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

) during startup for any lines containing

datacache

block_cache

to see if there are errors initializing the cache instance. References • using_starrocks/caching/block_cache.md • data_source/data_cache.md • deployment/shared_data/feature-support-shared-data.md • data_source/data_cache_troubleshooting.md

Mahesha Subrahamanya

07/31/2026, 4:29 AM

@Rocky In 4.1.1 shared-data CN mode, datacache_disk_size doesn't affect block_cache_disk_quota_bytes —what parameter controls it?

Rocky

07/31/2026, 4:30 AM

In StarRocks 4.1, the Data Cache (which includes what was previously called Block Cache) has been unified into a single instance. The behavior change you are seeing is likely due to the new Unified Data Cache being enabled by default (

datacache_unified_instance_enable=true

), which changes which parameters the system respects. In StarRocks 4.1.1 shared-data mode, the parameter

block_cache_disk_quota_bytes

in the metrics is controlled primarily by: 1. The
datacache_disk_size
Parameter In the unified instance,

datacache_disk_size

is the primary control. However, there are two critical things to note: * Default is 0: Unlike previous versions where some defaults were automatically applied, in the unified instance,

datacache_disk_size

defaults to

, which means disk caching is effectively disabled. * It supports Percentages: You can set it to a fixed size (e.g.,

500G

) or a percentage (e.g.,

80%

). * Per-Disk Limit: This value applies to each disk specified in your path configuration. If you have 2 disks and set

100G

, the total quota will be

200G

. 2. The
storage_root_path
Requirement For Compute Nodes (CN), the storage_root_path
is not set by default. If this parameter is empty or not configured in

cn.conf

, the Data Cache has no location to store data, and the quota will remain 0B/0B regardless of what you set in

datacache_disk_size

. * Fix: Ensure you have a path defined, e.g.,

storage_root_path = /data1/sr_cache;/data2/sr_cache

. 3. The
starlet_star_cache_disk_size_percent
Conflict In version 3.5, you likely used

starlet_star_cache_disk_size_percent

to control the cache. * In 4.1, this parameter is ignored if

datacache_unified_instance_enable

true

(which is the default). * To get the 3.5 behavior back, you should set

datacache_disk_size

to a percentage (e.g.,

80%

) instead of using the old starlet parameter. 4. Check for Auto-Adjustment If you have

datacache_auto_adjust_enable=true

(the default), the system might start with a 0 quota if it detects that the disk is already near the

disk_high_level

(default 90%). ────────── Recommended Configuration for `cn.conf`: To resolve the

0B/0B

issue, verify and set the following in your `cn.conf`:

Copy code

properties
# 1. Enable the cache
datacache_enable = true

# 2. Specify the directory (MANDATORY for CN)
storage_root_path = /your/fast/disk/path

# 3. Set the size (use a percentage to mimic 3.5 behavior)
datacache_disk_size = 80%

# 4. Optional: If you specifically want the old starlet logic (not recommended)
# datacache_unified_instance_enable = false

Verification After restarting the CN, run the following to see the active configuration:

Copy code

sql
-- Check if the system recognizes the path and size
SELECT * FROM information_schema.be_configs WHERE NAME IN ('storage_root_path', 'datacache_disk_size', 'datacache_enable');

-- Check the active metrics
SELECT * FROM information_schema.be_datacache_metrics;

DISK_QUOTA_BYTES

is still 0, check the CN leader log (

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

) for lines containing

fail to write block because no valid cache space

datacache

initialization errors, which usually point to permission issues or a missing

storage_root_path

. References • using_starrocks/caching/block_cache.md • data_source/data_cache.md

4 Views

Open in Slack

Previous Next