Elasticsearch init ops and kernel

[kvignesh1420]

This PR addresses the issue #1093 by adding implementations for the following:

• IO>ElasticsearchReadableInit Op.
• ElasticsearchReadableInitOp Kernel.
• tfio.experimental.elasticsearch.ElasticsearchIODataset template implementation and few helper methods.

Also, modified the BUILD file for Bazel build of elasticsearch_ops.

[kvignesh1420]

@yongtang the build on windows fails due to :

elasticsearch_ops.lo.lib(elasticsearch_kernels.obj) : error LNK2019: unresolved external symbol "public: __cdecl
 tensorflow::CurlHttpRequest::CurlHttpRequest(void)" (??0CurlHttpRequest@tensorflow@@QEAA@XZ) referenced in 
function "public: virtual class tensorflow::HttpRequest * __cdecl tensorflow::CurlHttpRequest::Factory::Create(void)" (?
Create@Factory@CurlHttpRequest@tensorflow@@UEAAPEAVHttpRequest@3@XZ)
2020-08-21T11:53:47.8604213Z bazel-out\x64_windows-fastbuild\bin\tensorflow_io\core\python\ops\libtensorflow_io.so : fatal error LNK1120: 1 unresolved externals
2020-08-21T11:53:47.9588587Z Target //tensorflow_io/core:python/ops/libtensorflow_io.so failed to build

It is unable to resolve CurlHttpRequest. Not sure why this is happening as the only other build using this class is core_ops and it seems to pass.

cc_library(
    name = "core_ops",
    srcs = [
        "kernels/archive_kernels.cc",
        "kernels/http_kernels.cc",
        "ops/core_ops.cc",
    ],
    copts = tf_io_copts(),
    linkstatic = True,
    deps = [
        ":dataset_ops",
        "@curl",
        "@libarchive",
    ],
    alwayslink = 1,
)

[yongtang]

@kvignesh1420 that was a known issue and has not been resolved yet. As you might see, core_ops is not enabled for windows build:

    ] + select({
        "@bazel_tools//src/conditions:windows": [],
        "//conditions:default": [
            "//tensorflow_io/core:core_ops",
            "//tensorflow_io/core:genome_ops",
            "//tensorflow_io/core:optimization",
            "//tensorflow_io/core:oss_ops",
            "//tensorflow_io/core/kernels/gsmemcachedfs:gs_memcached_file_system",
        ],
    }),

You can just move "//tensorflow_io/core:elasticsearch_ops", to the line below "//tensorflow_io/core:core_ops", and Windows will skip the elasticsearch op.

[kvignesh1420]

cc: @yongtang for review.

[yongtang]

LGTM

[yongtang]

@kvignesh1420 Thanks for the PR! Would you mind to add a test in a follow up PR?

[kvignesh1420]

@yongtang sure.

[kvignesh1420]

@yongtang can we return a tensor of tensors from the kernel to the OP?

For ex: The records I get from the api response are a list of jsons. In this scenario, I want to prepare a tensor for the column/field names and a tensor<tensor> where each inner tensor will contain the values for a column/field. The issue is, how do I return this tensor<tensor> to python from cpp kernels. Is there an alternative approach for these scenarios?

[yongtang]

@kvignesh1420 I would assume an easy way is to obtain the json string from dataset implementation, then using dataset.map() internally (could be part of the ElasticsearchIODataset class) to deserialize the json string into a structure of tensors. You can take a look at tfio.experimental.serialization.decode_json and see if this fits the need.

Let me know if you run into any issues.