[
https://issues.apache.org/jira/browse/GIRAPH-1136?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Nic Eggert updated GIRAPH-1136:
-------------------------------
Description:
>From the Worker Failure section in Chapter 6 of Practical Analytics with
>Apache Giraph:
{quote}
If checkpoints are enabled, you are guaranteed to have the initial graph data
safely stored in HDFS.
{quote}
Looking at the current code in
{{o.a.g.master.BspServiceMaster:getCheckpointStatus}}, it looks like this is no
longer the case. After a little bit of digging, it looks like this change was
introduced here:
https://github.com/apache/giraph/commit/5adca63deca25d84f4fdea053c35a85efc8bbb3d#diff-e16fdec9e3f573eba64cfe6eab512e19L657
(GIRAPH-933)
Was this change intentional? It seems like an initial checkpoint would be
desirable (I have a job where 1/3 of the runtime is spent loading input splits).
The simplest fix would be to just add a special case for Superstep 0. If that's
acceptable, I'd be happy to submit a PR.
was:
>From the Worker Failure section in Chapter 6 of Practical Analytics with
>Apache Giraph:
{quote}
If checkpoints are enabled, you are guaranteed to have the initial graph data
safely stored in HDFS.
{quote}
Looking at the current code in
{{o.a.g.master.BspServiceMaster:getCheckpointStatus}}, it looks like this is no
longer the case. After a little bit of digging, it looks like this change was
introduced here:
https://github.com/apache/giraph/commit/5adca63deca25d84f4fdea053c35a85efc8bbb3d#diff-e16fdec9e3f573eba64cfe6eab512e19L657
(GIRAPH-993)
Was this change intentional? It seems like an initial checkpoint would be
desirable (I have a job where 1/3 of the runtime is spent loading input splits).
The simplest fix would be to just add a special case for Superstep 0. If that's
acceptable, I'd be happy to submit a PR.
> Giraph no longer checkpoints after loading input
> ------------------------------------------------
>
> Key: GIRAPH-1136
> URL: https://issues.apache.org/jira/browse/GIRAPH-1136
> Project: Giraph
> Issue Type: Bug
> Components: bsp
> Affects Versions: 1.2.0
> Reporter: Nic Eggert
>
> From the Worker Failure section in Chapter 6 of Practical Analytics with
> Apache Giraph:
> {quote}
> If checkpoints are enabled, you are guaranteed to have the initial graph data
> safely stored in HDFS.
> {quote}
> Looking at the current code in
> {{o.a.g.master.BspServiceMaster:getCheckpointStatus}}, it looks like this is
> no longer the case. After a little bit of digging, it looks like this change
> was introduced here:
> https://github.com/apache/giraph/commit/5adca63deca25d84f4fdea053c35a85efc8bbb3d#diff-e16fdec9e3f573eba64cfe6eab512e19L657
> (GIRAPH-933)
> Was this change intentional? It seems like an initial checkpoint would be
> desirable (I have a job where 1/3 of the runtime is spent loading input
> splits).
> The simplest fix would be to just add a special case for Superstep 0. If
> that's acceptable, I'd be happy to submit a PR.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
