Re: Nodetool snapshot

[guo Maxwell]

yes you need to restore snapshot of the node’s own snapshot for every node

Abdul Patel 于2019年9月20日 周五上午2:08写道：

> Thanks , i guess i have both.
> So can we have either or?
> If i keep auto_snapshot? Can i remove nodetool snapshot?
> Woest case scenario if i wish to restore snapshot which one will be best
> option?
> Also if we restore snapshot do we need to have snapshot on all nodes?
>
>
> On Thursday, September 19, 2019, Jeff Jirsa  wrote:
>
>> You probably have auto_snapshot enabled, which takes snapshots when you
>> do certain things. You can disable that if you dont need it, but it
>> protects you against things like accidentally dropping / truncating a table.
>>
>> You may also be doing snapshots manually - if you do this, you can
>> 'nodetool clearsnapshot' to free up space.
>>
>>
>> On Thu, Sep 19, 2019 at 10:54 AM Abdul Patel  wrote:
>>
>>> Hey All,
>>>
>>> I found recentmy that the nodetool snapshot golder is creating almost
>>> 120GB of filea when my actual keyspace folder has 20GB only.
>>> Do we need to change any paramater to avoid this?
>>> Is this normal?
>>> I have 3.11.4 version
>>>
>> --
you are the apple of my eye !

Re: Nodetool snapshot

[Abdul Patel]

Thanks , i guess i have both.
So can we have either or?
If i keep auto_snapshot? Can i remove nodetool snapshot?
Woest case scenario if i wish to restore snapshot which one will be best
option?
Also if we restore snapshot do we need to have snapshot on all nodes?


On Thursday, September 19, 2019, Jeff Jirsa  wrote:

> You probably have auto_snapshot enabled, which takes snapshots when you do
> certain things. You can disable that if you dont need it, but it protects
> you against things like accidentally dropping / truncating a table.
>
> You may also be doing snapshots manually - if you do this, you can
> 'nodetool clearsnapshot' to free up space.
>
>
> On Thu, Sep 19, 2019 at 10:54 AM Abdul Patel  wrote:
>
>> Hey All,
>>
>> I found recentmy that the nodetool snapshot golder is creating almost
>> 120GB of filea when my actual keyspace folder has 20GB only.
>> Do we need to change any paramater to avoid this?
>> Is this normal?
>> I have 3.11.4 version
>>
>

Re: Nodetool snapshot

[Jeff Jirsa]

You probably have auto_snapshot enabled, which takes snapshots when you do
certain things. You can disable that if you dont need it, but it protects
you against things like accidentally dropping / truncating a table.

You may also be doing snapshots manually - if you do this, you can
'nodetool clearsnapshot' to free up space.


On Thu, Sep 19, 2019 at 10:54 AM Abdul Patel  wrote:

> Hey All,
>
> I found recentmy that the nodetool snapshot golder is creating almost
> 120GB of filea when my actual keyspace folder has 20GB only.
> Do we need to change any paramater to avoid this?
> Is this normal?
> I have 3.11.4 version
>

Nodetool snapshot

[Abdul Patel]

Hey All,

I found recentmy that the nodetool snapshot golder is creating almost 120GB
of filea when my actual keyspace folder has 20GB only.
Do we need to change any paramater to avoid this?
Is this normal?
I have 3.11.4 version

Re: Does cassandra serve reads during decommission process?

[Jeff Jirsa]

You can run removenode instead of decommission while it's down to avoid it
being online / serving reads at all.

You can also start cassandra with ` start_native_transport : false ` to
deter clients from connecting directly to it, though to be fair, that
doesnt eliminate the possibility that it's used for reads.



On Thu, Sep 19, 2019 at 10:20 AM Krishna PMV  wrote:

> Hi,
> I ran into a situation where a newly bootstrapped node in the cluster has
> crashed  (due to known issue) immediately after the bootstrap process and
> it remained dead for about 8 hours.
> Since the node is down for about 8 hours, its missing some data after I
> start cassandra. My application with queries of consistency LOCAL_ONE
> started failing as a result. Now, the node is in stopped state for about 12
> hours and I'm planning to decommission the node. (I know running repairs is
> another option but it will take a while since data size is about 350G and
> since it serves reads during repair process, I wanted to avoid this, if we
> can).
>
> So, I wanted to know if the node serves read requests during decommission
> process i.e., while leaving the cluster.
>
> Thanks!
>

Does cassandra serve reads during decommission process?

[Krishna PMV]

Hi,
I ran into a situation where a newly bootstrapped node in the cluster has
crashed  (due to known issue) immediately after the bootstrap process and
it remained dead for about 8 hours.
Since the node is down for about 8 hours, its missing some data after I
start cassandra. My application with queries of consistency LOCAL_ONE
started failing as a result. Now, the node is in stopped state for about 12
hours and I'm planning to decommission the node. (I know running repairs is
another option but it will take a while since data size is about 350G and
since it serves reads during repair process, I wanted to avoid this, if we
can).

So, I wanted to know if the node serves read requests during decommission
process i.e., while leaving the cluster.

Thanks!

Re: Cassandra Counters

[Federico Razzoli]

Hi Tarun,

That documentation page is a bit ambiguous. My understanding of it is that:

* Cassandra guarantees that counters are updated consistently across the
cluster by doing background reads, that don't affect write latency.
* If you use a consistency level stricter than ONE, the same read is done;
the difference is that it's not in background, and your write latency is
affected.
* They say that "typically" you want to use ONE because writes will be
faster, but don't specify what "typically" means in this case.

I suppose that "typically" means "if you are not concerned about Cassandra
nodes crashes". Here's my educated guess:

1. You write to node1, updating a counter. The new value should
theoretically be 10.
2. node1 returns the success.
3. node1 should read the new value in background, but it crashes before
finishing this read. What is the current state of the counter in node1?
>From the documentation, I fail to find this out.
4. You read the counter from node3 with ONE. You read 9.
5. You restart node1. I don't know what happens to the counter.

It would be great to have a clarification from Cassandra team.

Federico


On Wed, 18 Sep 2019 at 14:10, Tarun Chabarwal 
wrote:

> Hi
>
> I stumbled on this
> 
> post which says use consistency level ONE with counters. I'm using
> cassandra 3 with 3 copies in one data center. I've to support consistent
> reads.
>
> Can we do LOCAL_QUORUM read/write against counters ? Is there any downside
> of using quorum with counters ?
>
> If we can't use quorum then is it possible to get consistent reads ?
>
>
> Regards
> Tarun Chabarwal
>