Request with "parent+children" spans to the zipkin collector service fails with error: index out of range

[alexsuraykin]

If we send parent and children spans in request, jaeger will be failed with error: "runtime error: index out of range"

Steps to reproduce:

1. Run jaeger.
2. Send POST request to the "<jaeger_address>: <zipkin_collector_port>/api/v1/spans" with next json in the body:

[
    {
        "traceId": "4ed11df465275600",
        "id": "4ed11df465275600",
        "name": "get /",
        "timestamp": 1633073248674949,
        "duration": 14861,
        "localEndpoint": {
            "serviceName": "first_service",
            "ipv4": "10.0.2.15"
        },
        "annotations": [
            {
                "timestamp": 1633073248674949,
                "value": "sr"
            },
            {
                "timestamp": 163307324868981,
                "value": "ss"
            }
        ]
    },
    {
        "traceId": "4ed11df465275600",
        "parentId": "4ed11df465275600",
        "id": "c943743e25dc2cdf",
        "name": "get /api",
        "timestamp": 1633073248678309,
        "duration": 3360,
        "localEndpoint": {
            "serviceName": "first_service",
            "ipv4": "10.0.2.15"
        },
        "annotations": [
            {
                "timestamp": 1633073248678309,
                "value": "cs"
            },
            {
                "timestamp": 1633073248681669,
                "value": "sr"
            },
            {
                "timestamp": 1633073248685029,
                "value": "cr"
            },
            {
                "timestamp": 1633073248688388,
                "value": "ss"
            }
        ]
    }
]

Expected result: 202 Accepted
Actual result: 500 Internal Server Error

• Jaeger version: 1.28.0, master
• Deployment: Docker, Kubernetes

Additional context
During the debug, we found that the problem occurs in the SubmitZipkinBatch function.

After "Sanitize" there are "mSpans" array with 3 items and "bools" with labels of processed spans :

Where span with "parentId" field was converted in two items.

And then function creates "responses" array with length=2(from original "spans"), but iterates through "bools" with length =3. Eventually there is rises error "index out of range [2] with length 2".

[pavolloffay]

@alexsuraykin thanks for reporting this. Would you like to work on a fix?

[frzifus]

hey, out of interest, I went down the street and noticed this comment:

if the span is client and server we split it into two separate spans

Looking at the 2nd span, it matches this condition.

[rgroothuijsen]

The problem appears to be that the IsRPCClient() method only checks if the span has a kind field set to CLIENT, but this field was not introduced until v2. If I understand correctly, the v1 equivalent would be detecting the presence of a cs annotation.

[yurishkuro]

@alexsuraykin thanks for the reproducer. I think I have a fix, but I am curious how you came by such an input, because the second span is a dual client/server span which should not have been produced by the instrumentation, because it contains client-send an server-receive events that should be logged in different spans (that can share the ID). Are you migrating data from a Zipkin installation to Jaeger? (at query time Zipkin server may merge the two spans into one)