fix(opentelemetry): fix a bug that will cause a failure of sending tracing data to datadog agent

CHANGELOG/unreleased/kong/11599.yaml

@@ -0,0 +1,7 @@
+message: Fix a bug that will cause a failure of sending tracing data to datadog when value of x-datadog-parent-id header in requests is a short dec string
+type: bugfix
+scope: Core
+prs:
+  - 11599
+jiras:
+  - "FTI-5375"

kong/plugins/opentelemetry/otlp.lua

@@ -13,6 +13,7 @@ local table_merge = utils.table_merge
 local setmetatable = setmetatable
 
 local TRACE_ID_LEN = 16
+local SPAN_ID_LEN  = 8
 local NULL = "\0"
 local POOL_OTLP = "KONG_OTLP"
 local EMPTY_TAB = {}
@@ -74,17 +75,25 @@ local function transform_events(events)
   return pb_events
 end
 
--- translate the trace_id to otlp format
-local function to_ot_trace_id(trace_id)
-  local len = #trace_id
-  if len > TRACE_ID_LEN then
-    return trace_id:sub(-TRACE_ID_LEN)
+local function id_formatter(length)
+  return function(id)
+    local len = #id
+    if len > length then
+      return id:sub(-length)
 
-  elseif len < TRACE_ID_LEN then
-    return NULL:rep(TRACE_ID_LEN - len) .. trace_id
+    elseif len < length then
+      return NULL:rep(length - len) .. id
+    end
+
+    return id
   end
+end
 
-  return trace_id
+local to_ot_trace_id, to_ot_span_id
+do
+  -- translate the trace_id and span_id to otlp format
+  to_ot_trace_id = id_formatter(TRACE_ID_LEN)
+  to_ot_span_id  = id_formatter(SPAN_ID_LEN)
 end
 
 -- this function is to prepare span to be encoded and sent via grpc
@@ -96,7 +105,8 @@ local function transform_span(span)
     trace_id = to_ot_trace_id(span.trace_id),
     span_id = span.span_id,
     -- trace_state = "",
-    parent_span_id = span.parent_id or "",
+    parent_span_id = span.parent_id and
+                     to_ot_span_id(span.parent_id) or "",
     name = span.name,
     kind = span.kind or 0,
     start_time_unix_nano = span.start_time_ns,

spec/03-plugins/37-opentelemetry/01-otlp_spec.lua

@@ -119,4 +119,51 @@ describe("Plugin: opentelemetry (otlp)", function()
     end
   end)
 
+  it("check lengths of trace_id and span_id ", function ()
+    local TRACE_ID_LEN, PARENT_SPAN_ID_LEN = 16, 8
+    local default_span = {
+      name = "full-span",
+      trace_id = rand_bytes(16),
+      span_id = rand_bytes(8),
+      parent_id = rand_bytes(8),
+      start_time_ns = time_ns(),
+      end_time_ns = time_ns() + 1000,
+      should_sample = true,
+      attributes = {
+        foo = "bar",
+        test = true,
+        version = 0.1,
+      },
+      events = {
+        {
+          name = "evt1",
+          time_ns = time_ns(),
+          attributes = {
+            debug = true,
+          }
+        }
+      },
+    }
+
+    local test_spans = {}
+    local span1 = utils.deep_copy(default_span)
+    local span2 = utils.deep_copy(default_span)
+    span1.trace_id = rand_bytes(17)
+    span1.expected_tid = span1.trace_id:sub(-TRACE_ID_LEN)
+    span1.parent_id = rand_bytes(9)
+    span1.expected_pid = span1.parent_id:sub(-PARENT_SPAN_ID_LEN)
+    span2.trace_id = rand_bytes(15)
+    span2.expected_tid = '\0' .. span2.trace_id
+    span2.parent_id = rand_bytes(7)
+    span2.expected_pid = '\0' .. span2.parent_id
+    insert(test_spans, span1)
+    insert(test_spans, span2)
+
+    for _, span in ipairs(test_spans) do
+      local pb_span = otlp.transform_span(span)
+      assert(pb_span.parent_span_id == span.expected_pid)
+      assert(pb_span.trace_id == span.expected_tid)
+    end
+  end)
+
 end)