feat: add cached token metrics support for Amazon Bedrock

Author: poshinchen

src/strands/telemetry/metrics.py

@@ -11,7 +11,7 @@
 
 from ..telemetry import metrics_constants as constants
 from ..types.content import Message
-from ..types.streaming import Metrics, Usage
+from ..types.event_loop import Metrics, Usage
 from ..types.tools import ToolUse
 
 logger = logging.getLogger(__name__)
@@ -264,6 +264,21 @@ def update_usage(self, usage: Usage) -> None:
         self.accumulated_usage["outputTokens"] += usage["outputTokens"]
         self.accumulated_usage["totalTokens"] += usage["totalTokens"]
 
+        # Handle optional cached token metrics
+        if "cacheReadInputTokens" in usage:
+            cache_read_tokens = usage["cacheReadInputTokens"]
+            self._metrics_client.event_loop_cache_read_input_tokens.record(cache_read_tokens)
+            self.accumulated_usage["cacheReadInputTokens"] = (
+                self.accumulated_usage.get("cacheReadInputTokens", 0) + cache_read_tokens
+            )
+
+        if "cacheWriteInputTokens" in usage:
+            cache_write_tokens = usage["cacheWriteInputTokens"]
+            self._metrics_client.event_loop_cache_write_input_tokens.record(cache_write_tokens)
+            self.accumulated_usage["cacheWriteInputTokens"] = (
+                self.accumulated_usage.get("cacheWriteInputTokens", 0) + cache_write_tokens
+            )
+
     def update_metrics(self, metrics: Metrics) -> None:
         """Update the accumulated performance metrics with new metrics data.
 
@@ -325,11 +340,21 @@ def _metrics_summary_to_lines(event_loop_metrics: EventLoopMetrics, allowed_name
         f"├─ Cycles: total={summary['total_cycles']}, avg_time={summary['average_cycle_time']:.3f}s, "
         f"total_time={summary['total_duration']:.3f}s"
     )
-    yield (
-        f"├─ Tokens: in={summary['accumulated_usage']['inputTokens']}, "
-        f"out={summary['accumulated_usage']['outputTokens']}, "
-        f"total={summary['accumulated_usage']['totalTokens']}"
-    )
+
+    # Build token display with optional cached tokens
+    token_parts = [
+        f"in={summary['accumulated_usage']['inputTokens']}",
+        f"out={summary['accumulated_usage']['outputTokens']}",
+        f"total={summary['accumulated_usage']['totalTokens']}",
+    ]
+
+    # Add cached token info if present
+    if summary["accumulated_usage"].get("cacheReadInputTokens"):
+        token_parts.append(f"cache_read={summary['accumulated_usage']['cacheReadInputTokens']}")
+    if summary["accumulated_usage"].get("cacheWriteInputTokens"):
+        token_parts.append(f"cache_write={summary['accumulated_usage']['cacheWriteInputTokens']}")
+
+    yield f"├─ Tokens: {', '.join(token_parts)}"
     yield f"├─ Bedrock Latency: {summary['accumulated_metrics']['latencyMs']}ms"
 
     yield "├─ Tool Usage:"
@@ -421,6 +446,8 @@ class MetricsClient:
     event_loop_latency: Histogram
     event_loop_input_tokens: Histogram
     event_loop_output_tokens: Histogram
+    event_loop_cache_read_input_tokens: Histogram
+    event_loop_cache_write_input_tokens: Histogram
 
     tool_call_count: Counter
     tool_success_count: Counter
@@ -474,3 +501,9 @@ def create_instruments(self) -> None:
         self.event_loop_output_tokens = self.meter.create_histogram(
             name=constants.STRANDS_EVENT_LOOP_OUTPUT_TOKENS, unit="token"
         )
+        self.event_loop_cache_read_input_tokens = self.meter.create_histogram(
+            name=constants.STRANDS_EVENT_LOOP_CACHE_READ_INPUT_TOKENS, unit="token"
+        )
+        self.event_loop_cache_write_input_tokens = self.meter.create_histogram(
+            name=constants.STRANDS_EVENT_LOOP_CACHE_WRITE_INPUT_TOKENS, unit="token"
+        )

src/strands/telemetry/metrics_constants.py

@@ -13,3 +13,5 @@
 STRANDS_EVENT_LOOP_CYCLE_DURATION = "strands.event_loop.cycle_duration"
 STRANDS_EVENT_LOOP_INPUT_TOKENS = "strands.event_loop.input.tokens"
 STRANDS_EVENT_LOOP_OUTPUT_TOKENS = "strands.event_loop.output.tokens"
+STRANDS_EVENT_LOOP_CACHE_READ_INPUT_TOKENS = "strands.event_loop.cache_read.input.tokens"
+STRANDS_EVENT_LOOP_CACHE_WRITE_INPUT_TOKENS = "strands.event_loop.cache_write.input.tokens"

src/strands/types/event_loop.py

@@ -2,21 +2,25 @@
 
 from typing import Literal
 
-from typing_extensions import TypedDict
+from typing_extensions import Required, TypedDict
 
 
-class Usage(TypedDict):
+class Usage(TypedDict, total=False):
     """Token usage information for model interactions.
 
     Attributes:
-        inputTokens: Number of tokens sent in the request to the model..
+        inputTokens: Number of tokens sent in the request to the model.
         outputTokens: Number of tokens that the model generated for the request.
         totalTokens: Total number of tokens (input + output).
+        cacheReadInputTokens: Number of tokens read from cache (optional).
+        cacheWriteInputTokens: Number of tokens written to cache (optional).
     """
 
-    inputTokens: int
-    outputTokens: int
-    totalTokens: int
+    inputTokens: Required[int]
+    outputTokens: Required[int]
+    totalTokens: Required[int]
+    cacheReadInputTokens: int
+    cacheWriteInputTokens: int
 
 
 class Metrics(TypedDict):

tests/strands/event_loop/test_streaming.py

@@ -260,6 +260,18 @@ def test_extract_usage_metrics():
     assert tru_usage == exp_usage and tru_metrics == exp_metrics
 
 
+def test_extract_usage_metrics_with_cache_tokens():
+    event = {
+        "usage": {"inputTokens": 0, "outputTokens": 0, "totalTokens": 0, "cacheReadInputTokens": 0},
+        "metrics": {"latencyMs": 0},
+    }
+
+    tru_usage, tru_metrics = strands.event_loop.streaming.extract_usage_metrics(event)
+    exp_usage, exp_metrics = event["usage"], event["metrics"]
+
+    assert tru_usage == exp_usage and tru_metrics == exp_metrics
+
+
 @pytest.mark.parametrize(
     ("response", "exp_events"),
     [

tests/strands/telemetry/test_metrics.py

@@ -90,6 +90,7 @@ def usage(request):
         "inputTokens": 1,
         "outputTokens": 2,
         "totalTokens": 3,
+        "cacheWriteInputTokens": 10,
     }
     if hasattr(request, "param"):
         params.update(request.param)
@@ -315,17 +316,14 @@ def test_event_loop_metrics_update_usage(usage, event_loop_metrics, mock_get_met
         event_loop_metrics.update_usage(usage)
 
     tru_usage = event_loop_metrics.accumulated_usage
-    exp_usage = Usage(
-        inputTokens=3,
-        outputTokens=6,
-        totalTokens=9,
-    )
+    exp_usage = Usage(inputTokens=3, outputTokens=6, totalTokens=9, cacheWriteInputTokens=30)
 
     assert tru_usage == exp_usage
     mock_get_meter_provider.return_value.get_meter.assert_called()
     metrics_client = event_loop_metrics._metrics_client
     metrics_client.event_loop_input_tokens.record.assert_called()
     metrics_client.event_loop_output_tokens.record.assert_called()
+    metrics_client.event_loop_cache_write_input_tokens.record.assert_called()
 
 
 def test_event_loop_metrics_update_metrics(metrics, event_loop_metrics, mock_get_meter_provider):