incubator-cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Arindam Barua <aba...@247-inc.com>
Subject RE: gc_grace_seconds to 0?
Date Thu, 24 Oct 2013 01:05:52 GMT

Forwarding to the group in case this helps out anyone else.


>>If so, should I set gc_grace_seconds to a lower non-zero value like 1-2 days?
Yes.

A
-----------------
Aaron Morton
New Zealand
@aaronmorton

Co-Founder & Principal Consultant
Apache Cassandra Consulting
http://www.thelastpickle.com

On 23/10/2013, at 1:08 PM, Arindam Barua <abarua@247-inc.com<mailto:abarua@247-inc.com>>
wrote:


We are not doing deletes, but are setting ttls of 8 days on most of our columns (these are
not updates to existing columns). Hence it seems safe to reduce gc_grace_seconds to even 0
from a tombstones getting back to life concern. The motivation is that the lower gc_grace_seconds
will help us reclaim tombstoned data quickly, which really adds up otherwise since it will
otherwise linger around for an additional 10 days by default. However, I'm concerned about
the hints being ttl'd to 0, which would mean hints would be effectively turned off? If so,
should I set gc_grace_seconds to a lower non-zero value like 1-2 days?

Thanks,
Arindam

-----Original Message-----
From: Aaron Morton [mailto:aaron@thelastpickle.com<http://thelastpickle.com>]
Sent: Tuesday, October 22, 2013 11:50 AM
To: Arindam Barua
Subject: Re: Decommissioned nodes not leaving and Hinted Handoff flood


>>I haven't seen documentation for this elsewhere. I'm using Cassandra 1.1.5, and the
yaml seems to mentions "max_hint_window_in_ms" for that purpose.
max_hint_window_in_ms is how long we will collect hints for when a node is down.


>>We don't have any deletes in our application, and hence are considering making gc_grace_seconds
0. However, if this affects the ttls of hints, then we probably don't want it to be 0.
If you are not doing deletes you can leave gc_grace_seconds with the default.



thanks
aaron



-----------------
Aaron Morton
New Zealand
@aaronmorton

Co-Founder & Principal Consultant
Apache Cassandra Consulting
http://www.thelastpickle.com

On 23/10/2013, at 6:06 AM, Arindam Barua <abarua@247-inc.com<mailto:abarua@247-inc.com>>
wrote:



Hi Aaron,

If you have a chance, can you please respond to a question I had related to one of your emails
from earlier.


>>The hints are stored with a TTL that is the gc_grace_seconds for the CF a the time
the hint is written, so they will eventually be purged by compaction.
I haven't seen documentation for this elsewhere. I'm using Cassandra 1.1.5, and the yaml seems
to mentions "max_hint_window_in_ms" for that purpose.

We don't have any deletes in our application, and hence are considering making gc_grace_seconds
0. However, if this affects the ttls of hints, then we probably don't want it to be 0.

Thanks,
Arindam

[1]
hinted_handoff_enabled: true
# this defines the maximum amount of time a dead host will have hints
# generated.  After it has been dead this long, hints will be dropped.
max_hint_window_in_ms: 3600000 # one hour

From: Arindam Barua [mailto:abarua@247-inc.com]
Sent: Wednesday, October 16, 2013 8:32 PM
To: user@cassandra.apache.org
Subject: gc_grace_seconds to 0?


We don't do any deletes in our cluster, but do set ttls of 8 days on most of the columns.
After reading a bunch of earlier threads, I have concluded that I can safely set gc_grace_seconds
to 0 and not have to worry about expired columns coming back to life. However, I wanted to
know if there is any other downside to setting gc_grace_seconds to 0. Eg. I saw a mention
of the ttl of hints set to gc_grace_seconds.

Thanks,
Arindam

Mime
View raw message