[
https://issues.apache.org/jira/browse/BEAM-5404?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16623768#comment-16623768
]
Niel Markwick commented on BEAM-5404:
-------------------------------------
This bug is incorrect...
It was a bug in my testing which indicated that Java serialization is more
efficient
My mutations were 10 columns of 10K strings... but the values were _the same_
10K string.
ie: String stringValue = new String( /* 10K char array */)
Mutation m = Mutation.newInsertOrUpdateBuilder("table1")
.set("key").to(UUID.randomUUID().toString())
.set("value0").to(stringValue)
.set("value1").to(stringValue)
.set("value2").to(stringValue)
// etc
So when the custom serializer encoded this, it produced a ~100K byte array,
Java serialization was being clever: it only sees one String object to be
serialized and produced a ~10K byte array...
> Inefficient Serialization of Spanner MutationGroup in pipeline
> --------------------------------------------------------------
>
> Key: BEAM-5404
> URL: https://issues.apache.org/jira/browse/BEAM-5404
> Project: Beam
> Issue Type: Bug
> Components: io-java-gcp
> Affects Versions: 2.3.0, 2.4.0, 2.5.0, 2.6.0
> Reporter: Niel Markwick
> Assignee: Chamikara Jayalath
> Priority: Major
> Labels: pull-request-available
> Fix For: Not applicable
>
> Time Spent: 50m
> Remaining Estimate: 0h
>
> The Cloud Spanner connector uses a custom serialization mechanism to convert
> MutationGroup objects into a byte array.
> This mechanism is very inefficient producing byte arrays approx 10x larger
> than simple Java Serialization of the MutationGroup objects, which increases
> the resources needed by the connector to ~40x the size of the original
> mutations.
> There are no obvious benefits to using this custom serialization system, as
> the objects are deserialized within the pipeline itself.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
