You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@hive.apache.org by Tzur Turkenitz <tz...@gmail.com> on 2013/07/18 22:40:22 UTC
Hive Architecture - Execution on nodes
Hello,
Just finished reading the Hive-Architecture pdf, and failed to find the
answers I was hoping for. So here I am, hoping this community will shed some
light.
I think I know what the answers will be, I need that bolted down and
secured.
We are concerned on how data is transferred between data-nodes and hive,
especially when it comes to clusters were there's no SSL between nodes.
And this is the user-case:
1. Table employee is a Hive table, with SerDe
2. MapReduce job accesses the table Employees which holds Encrypted
data
3. SerDe decrypts the data
4. Post-SerDe output is returned to the MapReduce job and saved to a
new Hive table using a new Encryption implementation
The flow, as I think it currently is should be:
MapReduce Job -- > Read table metadata -- > SerDe creates map-reduce job --
> distributes across nodes
Which means that data is decrypted on the local nodes and then sent in
clear-text back to the original map-reduce job to be saved in a new table.
Is that correct? :(
RE: Hive Architecture - Execution on nodes
Posted by Tzur Turkenitz <tz...@gmail.com>.
Thank you Alan.
-----Original Message-----
From: Alan Gates [mailto:gates@hortonworks.com]
Sent: Thursday, July 18, 2013 5:45 PM
To: user@hive.apache.org
Subject: Re: Hive Architecture - Execution on nodes
On Jul 18, 2013, at 1:40 PM, Tzur Turkenitz wrote:
> Hello,
> Just finished reading the Hive-Architecture pdf, and failed to find the
answers I was hoping for. So here I am, hoping this community will shed some
light.
> I think I know what the answers will be, I need that bolted down and
secured.
>
> We are concerned on how data is transferred between data-nodes and hive,
especially when it comes to clusters were there's no SSL between nodes.
>
> And this is the user-case:
> 1. Table employee is a Hive table, with SerDe
> 2. MapReduce job accesses the table Employees which holds Encrypted
data
> 3. SerDe decrypts the data
> 4. Post-SerDe output is returned to the MapReduce job and saved to a
new Hive table using a new Encryption implementation
>
> The flow, as I think it currently is should be:
> MapReduce Job -- > Read table metadata -- > SerDe creates map-reduce job
-- > distributes across nodes
>
> Which means that data is decrypted on the local nodes and then sent in
clear-text back to the original map-reduce job to be saved in a new table.
> Is that correct? L
No. Data deserialization (which is what a serde does, not decryption) is
done as part of reading in the map reduce job. Mainly only query parsing,
validation, and planning is done on the client node.
Alan.
>
Re: Hive Architecture - Execution on nodes
Posted by Alan Gates <ga...@hortonworks.com>.
On Jul 18, 2013, at 1:40 PM, Tzur Turkenitz wrote:
> Hello,
> Just finished reading the Hive-Architecture pdf, and failed to find the answers I was hoping for. So here I am, hoping this community will shed some light.
> I think I know what the answers will be, I need that bolted down and secured.
>
> We are concerned on how data is transferred between data-nodes and hive, especially when it comes to clusters were there’s no SSL between nodes.
>
> And this is the user-case:
> 1. Table employee is a Hive table, with SerDe
> 2. MapReduce job accesses the table Employees which holds Encrypted data
> 3. SerDe decrypts the data
> 4. Post-SerDe output is returned to the MapReduce job and saved to a new Hive table using a new Encryption implementation
>
> The flow, as I think it currently is should be:
> MapReduce Job -- > Read table metadata -- > SerDe creates map-reduce job -- > distributes across nodes
>
> Which means that data is decrypted on the local nodes and then sent in clear-text back to the original map-reduce job to be saved in a new table.
> Is that correct? L
No. Data deserialization (which is what a serde does, not decryption) is done as part of reading in the map reduce job. Mainly only query parsing, validation, and planning is done on the client node.
Alan.
>