Mailing-List: contact hdfs-dev-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Reply-To: hdfs-dev@hadoop.apache.org
Received-SPF: pass (nike.apache.org: domain of atm@cloudera.com designates
 209.85.223.171 as permitted sender)
MIME-Version: 1.0
In-Reply-To: 
 <CAAu13zHYxe=wWQH19XdkeREyFoidKw=fKM3XEpg_tkA5Rt-n+w@mail.gmail.com>
References: 
 <CAAu13zHYxe=wWQH19XdkeREyFoidKw=fKM3XEpg_tkA5Rt-n+w@mail.gmail.com>
From: "Aaron T. Myers" <atm@cloudera.com>
Date: Sat, 20 Apr 2013 03:23:52 +0900
Message-ID: 
 <CA+4052nmKEYwZ0p4zSDzdQmjsYq5OXx-e2fQ1v6Q-iZ7xhoniQ@mail.gmail.com>
Subject: Re: Convention question for using DFSConfigKey constants : are zeros
 magic?
To: "hdfs-dev@hadoop.apache.org" <hdfs-dev@hadoop.apache.org>
Content-Type: multipart/alternative; boundary=089e013cbb8a037a9f04dabad4d0

--089e013cbb8a037a9f04dabad4d0
Content-Type: text/plain; charset=ISO-8859-1

Hi Jay,

On Sat, Apr 20, 2013 at 1:10 AM, Jay Vyas <jayunit100@gmail.com> wrote:

> I recently looked into the HDFS source tree to determine idioms with
> respect to a hairy debate about the threshold between what is and is not a
> magic number, and found that :  And it appears that the number zero is NOT
> considered magic - at least not in the HDFS source code.
>

It's certainly not magic in the Configuration class interpretation of it,
though I think if you surveyed the full source code you'd find that there
won't be much consistency with regard to ad hoc checks in the code for
certain values, like you've identified below.


>
> I found that that DFSConfigKeys.java defines DEFAULT values of zeros for
> some fields, and those defaults result in non-quantitative interpretation
> of the field.
>
> For example:
> dfs.image.transfer.bandwidthPerSec
>
> Is commented like so:
> public static final long DFS_IMAGE_TRANSFER_RATE_DEFAULT = 0;  //no
> throttling
>
> However in the implementation of these defaults, "magic" zeros are used
> without commenting:
> if (transferBandwidth > 0) {
>       throttler = new DataTransferThrottler(transferBandwidth);
> }
>
> --------
>
> Seems like the 0 above would be better replaced with
> DFS_IMAGE_TRANSFER_RATE_DEFAULT since the "no throttling" behaviour is
> defined with the constant in the DFSConfigKeys file, and not defined in the
>
> hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/GetImageServlet.java.
>

I don't agree with this. What if we later changed the default to something
greater than 0 in DFSConfigKeys? If the code were comparing against the
value DFS_IMAGE_TRANSFER_RATE_DEFAULT, the check in the code would then be
wrong. The only value for that config that should denote "no throttling" is
0, regardless of what the default is, so the explicit comparison against 0
makes sense to me.


>
>
> --------
>
> Trying to get a feel for if there are conventions  to enforce in some code
> reviews for our hadoop dependent configuration code.  We'd like to follow
> hadoopy idioms if possible..
>

I'd say the main conventions you should concern yourself with for this
purpose is config setting naming, e.g. use consistent prefixes within your
own code, use lower case separated by dots.and-dashes, etc.


>
>
> --
> Jay Vyas
> http://jayunit100.blogspot.com
>

--089e013cbb8a037a9f04dabad4d0--