Mailing-List: contact user-help@cassandra.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@cassandra.apache.org
Received-SPF: pass (athena.apache.org: domain of daniel.doubleday@gmx.net
 designates 213.165.64.23 as permitted sender)
From: Daniel Doubleday <daniel.doubleday@gmx.net>
Mime-Version: 1.0 (Apple Message framework v1084)
Content-Type: multipart/alternative; boundary=Apple-Mail-1-430113872
Subject: Re: Row cache
Date: Thu, 30 Jun 2011 18:03:15 +0200
In-Reply-To: 
 <OF6EC563D6.BC812CC2-ONC22578BF.003DF191-C22578BF.0040C0DF@il.ibm.com>
To: user@cassandra.apache.org
References: 
 <OF6EC563D6.BC812CC2-ONC22578BF.003DF191-C22578BF.0040C0DF@il.ibm.com>
Message-Id: <8A7E912E-9429-4FF2-8EA7-6FD4C2E5BA26@gmx.net>


--Apple-Mail-1-430113872
Content-Transfer-Encoding: quoted-printable
Content-Type: text/plain;
	charset=us-ascii

Here's my understanding of things ... (this applies only for the regular =
heap implementation of row cache)

> Why Cassandra does not cache a row that was requested few times?=20

What does the cache capacity read. Is it > 0?

> What the ReadCount attribute in ColumnFamilies indicates and why it =
remains zero.=20

Hm I had that too one time (read count wont go up while there were =
reads). But I didn't have the time to debug.

> How can I know from where Cassandra read a row (from MEMTable,RowCache =
or SSTable)?=20

It will always read from=20
row cache or=20
memtable(s) and sstable(s)

jmx should tell you (hits go up)

> does the following correct? In read operation Cassandra looks for the =
row in the MEMTable - if not found it looks in the row-cache - if not =
found it looks in SSTable (after looking in the key-cache to optimize =
the access to the SSTable)?=20

No.=20

If row cache capacity is > 0 then a read will check if the row is in =
cache if not it read the entire row and cache it. Then / or if row was =
in cache already it will read from there and apply the respective filter =
to the cached CF.   Writes update memtable and row cache when the row is =
cached. I must admit that I still dont quite understand why there's no =
race here. I haven't found any cache lock. So someone else should =
explain why a concurrent read / write cannot produce a lost update in =
the cached row.

If capacity is 0 then it will read from the current memtable, the =
memtable(s) that are being flushed and all sstables that may contain the =
row (filtered by bloom filter)

Hope that's correct and helps.

Cheers,
Daniel


--Apple-Mail-1-430113872
Content-Transfer-Encoding: 7bit
Content-Type: text/html;
	charset=us-ascii

<html><head></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; ">Here's my understanding of things ... (this applies only for the regular heap implementation of row cache)<div><br><div><blockquote type="cite"><font size="2" face="sans-serif">Why Cassandra does not cache a row that
was requested few times?</font>
<br></blockquote><div><br></div>What does the cache capacity read. Is it &gt; 0?</div><div><br><blockquote type="cite"><font size="2" face="sans-serif">What the ReadCount attribute in ColumnFamilies
indicates and why it remains zero.</font>
<br></blockquote><div><br></div>Hm I had that too one time (read count wont go up while there were reads). But I didn't have the time to debug.</div><div><br><blockquote type="cite"><font size="2" face="sans-serif">How can I know from where Cassandra
read a row (from MEMTable,RowCache or SSTable)?</font>
<br></blockquote><div><br></div><div>It will always read from&nbsp;</div><div>row cache or&nbsp;</div><div>memtable(s) and sstable(s)</div><div><br></div><div>jmx should tell you (hits go up)</div><br><blockquote type="cite"><font size="2" face="sans-serif">does the following correct? In read
operation Cassandra looks for the row in the MEMTable - if not found it
looks in the row-cache - if not found it looks in SSTable (after looking
in the key-cache to optimize the access to the SSTable)? </font>
<br></blockquote><div><br></div>No.&nbsp;</div><div><br></div><div>If row cache capacity is &gt; 0 then a read will check if the row is in cache if not it read the entire row and cache it. Then / or if row was in cache already it will read from there and apply the respective filter to the cached CF. &nbsp; Writes update memtable and row cache when the row is cached. I must admit that I still dont quite understand why there's no race here. I haven't found any cache lock. So someone else should explain why a concurrent read / write cannot produce a lost update in the cached row.</div><div><br></div><div>If capacity is 0 then it will read from the current memtable, the memtable(s) that are being flushed and all sstables that may contain the row (filtered by bloom filter)</div><div><br></div><div>Hope that's correct and helps.</div><div><br></div><div>Cheers,</div><div>Daniel</div><br></div></body></html>
--Apple-Mail-1-430113872--