vranganathan avatar image
vranganathan asked Erick Ramirez edited

nodetool load does not seem to refresh


UN  172.x.x.x  186.44 TB  256          ?       f2f425ce-1059-4ef3-b1d4-e095701a9439  us-east-1e


*Restart the node*



UN  172.x.x.x  231.59 GB  256          ?       f2f425ce-1059-4ef3-b1d4-e095701a9439  us-east-1e

How is the load on a host decided? Official documentation says

The amount of file system data under the cassandra data directory after excluding all content in the snapshots subdirectories. Because all SSTable data files are included, any data that is not cleaned up, such as TTL-expired cell or tombstoned data) is counted.

This number (186.44 TB) did not come down after a "nodetool cleanup" / after the node is done with all the compactions.

This only came down after a restart. Why does this happen?

10 |1000

Up to 8 attachments (including images) can be used with a maximum of 1.0 MiB each and 10.0 MiB total.

1 Answer

aholmber avatar image
aholmber answered aholmber commented

Did you notice that those numbers have different units? It looks like it has gone down three orders of magnitude from hundreds of terabytes to a couple hundred gigabytes.

2 comments Share
10 |1000

Up to 8 attachments (including images) can be used with a maximum of 1.0 MiB each and 10.0 MiB total.

vranganathan avatar image vranganathan commented ·

@aholmber Yes, I am aware of it and that is exactly my question. How come restarting a node clears up so much load ?

Little more context:
I did a nodetool cleanup and waited for all the active compactions to finish. After doing this, I waited for 3 hrs and then did this comparison. (Ran nodetool status before restart and immediately after restart)

This makes me think that the output nodetool status gives out might be flawed. There was no data loss, everything seems just perfect except this huge reduction in the `load`.

0 Likes 0 ·
aholmber avatar image aholmber commented ·

I see I misunderstood what was being asked. You're specifically wondering why the cleanup is not reflected until after the node is restarted.

I don't know if this is by design, but it does seem surprising to me. If it's reproducible, they might appreciate a ticket on the Apache Cassandra (TM) Jira.

0 Likes 0 ·