fix(plugin): OTEL exporting with translate (#10332)

Co-authored-by: Chrono <chrono_cpp@me.com>
Co-authored-by: Michael Martin <3277009+flrgh@users.noreply.github.com>

CHANGELOG.md

@@ -195,6 +195,8 @@
     [#10160](https://github.com/Kong/kong/pull/10160)
   - For `http.flavor`. It should be a string value, not a double.
     [#10160](https://github.com/Kong/kong/pull/10160)
+- **OpenTelemetry**: Fix a bug that when getting the trace of other formats, the trace ID reported and propagated could be of incorrect length.
+    [#10332](https://github.com/Kong/kong/pull/10332)
 - **OAuth2**: `refresh_token_ttl` is now limited between `0` and `100000000` by schema validator. Previously numbers that are too large causes requests to fail.
   [#10068](https://github.com/Kong/kong/pull/10068)
 

kong/plugins/opentelemetry/handler.lua

@@ -19,7 +19,8 @@ local propagation_parse = propagation.parse
 local propagation_set = propagation.set
 local null = ngx.null
 local encode_traces = otlp.encode_traces
-local transform_span = otlp.transform_span
+local translate_span_trace_id = otlp.translate_span
+local encode_span = otlp.transform_span
 
 local _log_prefix = "[otel] "
 
@@ -108,7 +109,7 @@ local function process_span(span, queue)
     span.trace_id = trace_id
   end
 
-  local pb_span = transform_span(span)
+  local pb_span = encode_span(span)
 
   queue:add(pb_span)
 end
@@ -132,6 +133,7 @@ function OpenTelemetryHandler:access()
   end
 
   -- overwrite trace id
+  -- as we are in a chain of existing trace
   if trace_id then
     root_span.trace_id = trace_id
     kong.ctx.plugin.trace_id = trace_id
@@ -145,7 +147,7 @@ function OpenTelemetryHandler:access()
     root_span.parent_id = parent_id
   end
 
-  propagation_set("preserve", header_type, root_span, "w3c")
+  propagation_set("preserve", header_type, translate_span_trace_id(root_span), "w3c")
 end
 
 function OpenTelemetryHandler:log(conf)

kong/plugins/opentelemetry/otlp.lua

@@ -10,8 +10,13 @@ local insert = table.insert
 local tablepool_fetch = tablepool.fetch
 local tablepool_release = tablepool.release
 local deep_copy = utils.deep_copy
+local shallow_copy = utils.shallow_copy
 local table_merge = utils.table_merge
+local getmetatable = getmetatable
+local setmetatable = setmetatable
 
+local TRACE_ID_LEN = 16
+local NULL = "\0"
 local POOL_OTLP = "KONG_OTLP"
 local EMPTY_TAB = {}
 
@@ -72,9 +77,43 @@ local function transform_events(events)
   return pb_events
 end
 
+-- translate the span to otlp format
+-- currently it only adjust the trace id to 16 bytes
+-- we don't do it in place because the span is shared
+-- and we need to preserve the original information as much as possible
+local function translate_span_trace_id(span)
+  local trace_id = span.trace_id
+  local len = #trace_id
+  local new_id = trace_id
+
+  if len == TRACE_ID_LEN then
+    return span
+  end
+
+  -- make sure the trace id is of 16 bytes
+  if len > TRACE_ID_LEN then
+    new_id = trace_id:sub(-TRACE_ID_LEN)
+
+  elseif len < TRACE_ID_LEN then
+    new_id = NULL:rep(TRACE_ID_LEN - len) .. trace_id
+  end
+
+  local translated = shallow_copy(span)
+  setmetatable(translated, getmetatable(span))
+
+  translated.trace_id = new_id
+  span = translated
+
+  return span
+end
+
+-- this function is to prepare span to be encoded and sent via grpc
+-- TODO: renaming this to encode_span
 local function transform_span(span)
   assert(type(span) == "table")
 
+  span = translate_span_trace_id(span)
+
   local pb_span = {
     trace_id = span.trace_id,
     span_id = span.span_id,
@@ -161,6 +200,7 @@ do
 end
 
 return {
+  translate_span = translate_span_trace_id,
   transform_span = transform_span,
   encode_traces = encode_traces,
 }

spec/03-plugins/37-opentelemetry/03-propagation_spec.lua

@@ -160,5 +160,27 @@ describe("propagation tests #" .. strategy, function()
 
     assert.equals(trace_id, json.headers["ot-tracer-traceid"])
   end)
+
+  it("propagates dd headers", function()
+    local trace_id = gen_trace_id()
+    local trace_id_truncated = trace_id:sub(1, 16)
+    local span_id = gen_span_id()
+    local r = proxy_client:get("/", {
+      headers = {
+        ["ot-tracer-traceid"] = trace_id_truncated,
+        ["ot-tracer-spanid"] = span_id,
+        ["ot-tracer-sampled"] = "1",
+        host = "http-route",
+      },
+    })
+    local body = assert.response(r).has.status(200)
+    local json = cjson.decode(body)
+
+    assert.equals(#trace_id, #json.headers["ot-tracer-traceid"],
+                  "trace ID was not padded correctly")
+
+    local expected = string.rep("0", 16) .. trace_id_truncated
+    assert.equals(expected, json.headers["ot-tracer-traceid"])
+  end)
 end)
 end