We spin up our slaves first, then the Hadoop namenode + Accumulo
master afterward. By the time that comes up, tserver has exited
on the Accumulo slaves.

Is there a way to increase the tserver effort for connection
to the master?

Aside: is there really any reason for tserver to exit? Why not
just do an exponential backoff of connection attempts until
30 seconds is hit, and then stay there infinitely?

Not sure if non- existence of the name node is something we want to deal
with. I worry this will cause confusion and ambiguity for first time users
who point Accumulo at an empty Hadoop configuration.

Sent from my phone, please pardon the typos and brevity.
On Feb 27, 2013 9:39 AM, "Jeff Blaine" <[EMAIL PROTECTED]> wrote:

> We spin up our slaves first, then the Hadoop namenode + Accumulo
> master afterward. By the time that comes up, tserver has exited
> on the Accumulo slaves.
>
> Is there a way to increase the tserver effort for connection
> to the master?
>
> Aside: is there really any reason for tserver to exit? Why not
> just do an exponential backoff of connection attempts until
> 30 seconds is hit, and then stay there infinitely?
>

We already wait on the NN.

-Eric

On Wed, Feb 27, 2013 at 10:06 AM, John Vines <[EMAIL PROTECTED]> wrote:

> Not sure if non- existence of the name node is something we want to deal
> with. I worry this will cause confusion and ambiguity for first time users
> who point Accumulo at an empty Hadoop configuration.
>
> Sent from my phone, please pardon the typos and brevity.
> On Feb 27, 2013 9:39 AM, "Jeff Blaine" <[EMAIL PROTECTED]> wrote:
>
>> We spin up our slaves first, then the Hadoop namenode + Accumulo
>> master afterward. By the time that comes up, tserver has exited
>> on the Accumulo slaves.
>>
>> Is there a way to increase the tserver effort for connection
>> to the master?
>>
>> Aside: is there really any reason for tserver to exit? Why not
>> just do an exponential backoff of connection attempts until
>> 30 seconds is hit, and then stay there infinitely?
>>
>

Wait for it to become extant or just wait for it to exit safe mode?
On Wed, Feb 27, 2013 at 10:13 AM, Eric Newton <[EMAIL PROTECTED]> wrote:

> We already wait on the NN.
>
> -Eric
>
>
>
> On Wed, Feb 27, 2013 at 10:06 AM, John Vines <[EMAIL PROTECTED]> wrote:
>
>> Not sure if non- existence of the name node is something we want to deal
>> with. I worry this will cause confusion and ambiguity for first time users
>> who point Accumulo at an empty Hadoop configuration.
>>
>> Sent from my phone, please pardon the typos and brevity.
>> On Feb 27, 2013 9:39 AM, "Jeff Blaine" <[EMAIL PROTECTED]> wrote:
>>
>>> We spin up our slaves first, then the Hadoop namenode + Accumulo
>>> master afterward. By the time that comes up, tserver has exited
>>> on the Accumulo slaves.
>>>
>>> Is there a way to increase the tserver effort for connection
>>> to the master?
>>>
>>> Aside: is there really any reason for tserver to exit? Why not
>>> just do an exponential backoff of connection attempts until
>>> 30 seconds is hit, and then stay there infinitely?
>>>
>>
>

See Accumulo.waitForZookeeperAndHdfs(FileSystem fs)

-Eric
On Wed, Feb 27, 2013 at 10:23 AM, John Vines <[EMAIL PROTECTED]> wrote:

> Wait for it to become extant or just wait for it to exit safe mode?
>
>
> On Wed, Feb 27, 2013 at 10:13 AM, Eric Newton <[EMAIL PROTECTED]>wrote:
>
>> We already wait on the NN.
>>
>> -Eric
>>
>>
>>
>> On Wed, Feb 27, 2013 at 10:06 AM, John Vines <[EMAIL PROTECTED]> wrote:
>>
>>> Not sure if non- existence of the name node is something we want to deal
>>> with. I worry this will cause confusion and ambiguity for first time users
>>> who point Accumulo at an empty Hadoop configuration.
>>>
>>> Sent from my phone, please pardon the typos and brevity.
>>> On Feb 27, 2013 9:39 AM, "Jeff Blaine" <[EMAIL PROTECTED]> wrote:
>>>
>>>> We spin up our slaves first, then the Hadoop namenode + Accumulo
>>>> master afterward. By the time that comes up, tserver has exited
>>>> on the Accumulo slaves.
>>>>
>>>> Is there a way to increase the tserver effort for connection
>>>> to the master?
>>>>
>>>> Aside: is there really any reason for tserver to exit? Why not
>>>> just do an exponential backoff of connection attempts until
>>>> 30 seconds is hit, and then stay there infinitely?
>>>>
>>>
>>
>

Sure, just create a ticket and submit a patch, if you're up to it.

-Eric
On Wed, Feb 27, 2013 at 9:38 AM, Jeff Blaine <[EMAIL PROTECTED]> wrote:

> We spin up our slaves first, then the Hadoop namenode + Accumulo
> master afterward. By the time that comes up, tserver has exited
> on the Accumulo slaves.
>
> Is there a way to increase the tserver effort for connection
> to the master?
>
> Aside: is there really any reason for tserver to exit? Why not
> just do an exponential backoff of connection attempts until
> 30 seconds is hit, and then stay there infinitely?
>

On 2/27/2013 9:48 AM, Eric Newton wrote:
> Sure, just create a ticket and submit a patch, if you're up to it.

I'm not afraid of contributing here. I do elsewhere. Unfortunately,
a patch + custom build for our project is not allowed. The effort
requires unpatched 1.4.2, so I figured I'd ask.

It looks like (after browsing the source a bit), I will have to
figure out a workaround.

If I'm feeling saucy and can make sense of it, I will try to
follow up with a patch some day so that this is addressed for
the next person. I'm not a Java developer, but do comprehend
most of the basic code I've seen in the repo based on previous
Java reading.

> On Wed, Feb 27, 2013 at 9:38 AM, Jeff Blaine <[EMAIL PROTECTED]
> <mailto:[EMAIL PROTECTED]>> wrote:
>
>     We spin up our slaves first, then the Hadoop namenode + Accumulo
>     master afterward. By the time that comes up, tserver has exited
>     on the Accumulo slaves.
>
>     Is there a way to increase the tserver effort for connection
>     to the master?
>
>     Aside: is there really any reason for tserver to exit? Why not
>     just do an exponential backoff of connection attempts until
>     30 seconds is hit, and then stay there infinitely?
>
>

Turns out I am hitting a bug (as far as I can tell) that is fixed in
trunk:

http://svn.apache.org/viewvc/accumulo/trunk/server/src/main/java/org/apache/accumulo/server/Accumulo.java?r1=1437810&r2=1438335&diff_format=h

I believe the prior code is trying to talk to Zookeeper before it
has called Accumulo.waitForZookeeperAndHdfs().  No?

Seems this was fixed for a totally different reason (ACCUMULO-928,
which doesn't talk about what I am talking about specifically)
and unknowingly addressed another bug.

On 2/27/2013 11:38 AM, Jeff Blaine wrote:
> On 2/27/2013 9:48 AM, Eric Newton wrote:
>> Sure, just create a ticket and submit a patch, if you're up to it.
>
> I'm not afraid of contributing here. I do elsewhere. Unfortunately,
> a patch + custom build for our project is not allowed. The effort
> requires unpatched 1.4.2, so I figured I'd ask.
>
> It looks like (after browsing the source a bit), I will have to
> figure out a workaround.
>
> If I'm feeling saucy and can make sense of it, I will try to
> follow up with a patch some day so that this is addressed for
> the next person. I'm not a Java developer, but do comprehend
> most of the basic code I've seen in the repo based on previous
> Java reading.
>
>> On Wed, Feb 27, 2013 at 9:38 AM, Jeff Blaine <[EMAIL PROTECTED]
>> <mailto:[EMAIL PROTECTED]>> wrote:
>>
>>     We spin up our slaves first, then the Hadoop namenode + Accumulo
>>     master afterward. By the time that comes up, tserver has exited
>>     on the Accumulo slaves.
>>
>>     Is there a way to increase the tserver effort for connection
>>     to the master?
>>
>>     Aside: is there really any reason for tserver to exit? Why not
>>     just do an exponential backoff of connection attempts until
>>     30 seconds is hit, and then stay there infinitely?
>>
>>