ignite-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Raymond Wilson <raymond_wil...@trimble.com>
Subject RE: Specifying location of persistent storage location
Date Tue, 05 Sep 2017 08:17:46 GMT
I wasn’t thinking of a snapshot so much, as a situation where a server
(particularly a virtual server) that might need to be bounced or replaced
during a maintenance window. If this was on AWS the process is pretty
simple – create a new server (for example with an updated patched AMI) and
reattach the EBS volume containing the persisted data. This server may well
not have the same IP address (though that could be mitigated with a DNS
name).



In any event, I’m just trying to convey a sense of unease over how IP
configuration controls the naming of the location of persistent data on an
Ignite server node and how (especially with a default configuration) this
can result in unpredictable changes of the name of the folder Ignite
expects to find its persisted data.



*From:* Dmitriy Setrakyan [mailto:dsetrakyan@apache.org]
*Sent:* Tuesday, September 5, 2017 6:12 PM
*To:* user <user@ignite.apache.org>
*Subject:* Re: Specifying location of persistent storage location







On Mon, Sep 4, 2017 at 8:40 PM, Raymond Wilson <raymond_wilson@trimble.com>
wrote:

Thanks.



I get the utility of specifying the network address to bind to; I’m not
convinced using that to derive the name of the internal data store is a
good idea! J

 For instance, what if you have to move a persistent data store to a
different server? Or are you saying everybody sets LocalHost or 120.0.0.1
to ensure the folder name is always essentially local host?



I think what you are asking about is a database backup or a snapshot.
Ignite does not support it out of the box, but you may wish to look at the
3rd party solutions, e.g. the one provided by GridGain -
https://docs.gridgain.com/docs/data-snapshots







*From:* Dmitriy Setrakyan [mailto:dsetrakyan@apache.org]
*Sent:* Tuesday, September 5, 2017 3:09 PM
*To:* user <user@ignite.apache.org>


*Subject:* Re: Specifying location of persistent storage location







On Mon, Sep 4, 2017 at 6:07 PM, Raymond Wilson <raymond_wilson@trimble.com>
wrote:

Dmitriy,



I set up an XML file based on the default one and added the two elements
you noted.



However, this has brought up an issue in that the XML file and an
IgniteConfiguration instance can’t both be provided to the Ignition.Start()
call. So I changed it to use the DiscoverSPI aspect of IgniteConfiguration
and set LocalAddress to “127.0.0.1” and LocalPort to 47500.



This did change the name of the persistence folder to be “127_0_0_1_47500”
as you suggested.



While this resolves my current issue with the folder name changing, it
still seems fragile as network configuration aspects of the server Ignite
is running on have a direct impact on an internal aspect of its
configuration (ie: the location where to store the persisted data). A DHCP
IP lease renewal or an internal DNS domain change or an internal IT
department change to using IPv6 addressing (among other things) could cause
problems when a node restarts and decides the location of its data is
different.



Do you know how GridGain manage this in their enterprise deployments using
persistence?



I am glad the issue is resolved. By default, Ignite will bind to all the
local network interfaces, and if they are provided in different order, it
may create the situation you witnessed.



All enterprise users explicitly specify which network address to bind to,
just like you did. This helps avoid any kind of magic in production.









Thanks,
Raymond.



*From:* Dmitriy Setrakyan [mailto:dsetrakyan@apache.org]
*Sent:* Tuesday, September 5, 2017 11:41 AM


*To:* user <user@ignite.apache.org>
*Cc:* Raymond Wilson <raymond_wilson@trimble.com>
*Subject:* Re: Specifying location of persistent storage location





On Mon, Sep 4, 2017 at 4:28 PM, Raymond Wilson <raymond_wilson@trimble.com>
wrote:

Hi,



It’s possible this could cause change in the folder name, though I do not
think this is an issue in my case. Below are three different folder names I
have seen. All use the same port number, but differ in terms of the IPV6
address (I have also seen variations where the IPv6 address is absent in
the folder name).

0_0_0_0_0_0_0_1_10_0_75_1_10_3_72_117_127_0_0_1_192_168_121_1_192_168_178_27_192_168_3_1_2406_e007_9e5_1_9cc8_92bc_50c9_6794_2406_e007_9e5_1_c5d8_af4b_55b2_582a_47500
,

0_0_0_0_0_0_0_1_10_0_75_1_10_3_72_117_127_0_0_1_192_168_121_1_192_168_178_27_192_168_3_1_2406_e007_9e5_1_a58c_2f32_8005_b03d_2406_e007_9e5_1_c5d8_af4b_55b2_582a_47500

0_0_0_0_0_0_0_1_10_0_75_1_10_3_72_117_127_0_0_1_192_168_121_1_192_168_178_27_192_168_3_1_2406_e007_38b4_1_858c_f0ab_bc60_54ab_2406_e007_38b4_1_c5d8_af4b_55b2_582a_47500



I start the nodes in my local setup in a well defined order so I would
expect the port to be the same. I did once start a second instance by
mistake and did see the port number incremented in the folder name.



Are you suggesting the two changes you note below will result in the same
folder name being chosen every time, unlike above?





Yes, exactly. My suggestions will ensure that you explicitly bind to the
same address every time.











Thanks,

Raymond.



*From:* Dmitriy Setrakyan [mailto:dsetrakyan@apache.org]
*Sent:* Tuesday, September 5, 2017 11:17 AM
*To:* user <user@ignite.apache.org>
*Cc:* Raymond Wilson <raymond_wilson@trimble.com>
*Subject:* Re: Specifying location of persistent storage location







On Mon, Sep 4, 2017 at 3:37 PM, Raymond Wilson <raymond_wilson@trimble.com>
wrote:

Hi,



I definitely have not had more than one server node running at the same
time (though there have been more than one client node running on the same
machine).



I suspect what is happening is that one or more of the network interfaces
on the machine can have their address change dynamically. What I thought of
as a GUID is actually (I think) an IPv6 address attached to one of the
interfaces. This aspect of the folder name tends to come and go.



You can see from the folder names below that there are quite a number of
addresses involved. This seems to be fragile (and I certainly see the name
of this folder changing frequently), so I think being able to set it to
something concrete would be a good idea.





I think I understand what is happening. Ignite starts off with a default
port, and then starts incrementing it with every new node started on the
same host. Perhaps you start server and client nodes in different order
sometimes which causes server to bind to a different port.



To make sure that your server node binds to the same port all the time, you
should try specifying it explicitly in the server node configuration, like
so (forgive me if this snippet does not compile):







*<property name="discoverySpi">         <bean
class="org.apache.ignite.spi.discovery.tcp.TcpDiscoverySpi">
 <property name="localPort" value="40123"/>         </bean></property>*



Please make sure that the client nodes either don't have any port
configured, or have a different port configured.



You should also make sure that Ignite always binds to the desired local
interface on client and server nodes, by specifying
IgniteConfiguration.setLocalHost(...) property, or like so in XML:



*<property name="localHost" value="my.local.ip.address"/>*



If my theory is correct, Ignite should make sure that the clients and
servers cannot theoretically bind to the same port. I will double check it
with the community and file a ticket if needed.

Mime
View raw message