holmesgpt 0.15.0__py3-none-any.whl → 0.16.0__py3-none-any.whl

holmes/__init__.py

@@ -1,5 +1,5 @@
 # This is patched by github actions during release
-__version__ = "0.15.0"
+__version__ = "0.16.0"
 
 # Re-export version functions from version module for backward compatibility
 from .version import (

holmes/common/env_vars.py

@@ -100,4 +100,12 @@ MAX_EVIDENCE_DATA_CHARACTERS_BEFORE_TRUNCATION = int(
     os.environ.get("MAX_EVIDENCE_DATA_CHARACTERS_BEFORE_TRUNCATION", 3000)
 )
 
+ENABLE_CONVERSATION_HISTORY_COMPACTION = load_bool(
+    "ENABLE_CONVERSATION_HISTORY_COMPACTION", default=True
+)
+
 DISABLE_PROMETHEUS_TOOLSET = load_bool("DISABLE_PROMETHEUS_TOOLSET", False)
+
+RESET_REPEATED_TOOL_CALL_CHECK_AFTER_COMPACTION = load_bool(
+    "RESET_REPEATED_TOOL_CALL_CHECK_AFTER_COMPACTION", True
+)

holmes/core/llm.py

@@ -26,6 +26,8 @@ from holmes.common.env_vars import (
     ROBUSTA_API_ENDPOINT,
     THINKING,
     EXTRA_HEADERS,
+    TOOL_MAX_ALLOCATED_CONTEXT_WINDOW_PCT,
+    TOOL_MAX_ALLOCATED_CONTEXT_WINDOW_TOKENS,
 )
 from holmes.core.supabase_dal import SupabaseDal
 from holmes.utils.env import environ_get_safe_int, replace_env_vars_values
@@ -41,6 +43,13 @@ MODEL_LIST_FILE_LOCATION = os.environ.get(
 
 OVERRIDE_MAX_OUTPUT_TOKEN = environ_get_safe_int("OVERRIDE_MAX_OUTPUT_TOKEN")
 OVERRIDE_MAX_CONTENT_SIZE = environ_get_safe_int("OVERRIDE_MAX_CONTENT_SIZE")
+
+
+def get_context_window_compaction_threshold_pct() -> int:
+    """Get the compaction threshold percentage at runtime to support test overrides."""
+    return environ_get_safe_int("CONTEXT_WINDOW_COMPACTION_THRESHOLD_PCT", default="95")
+
+
 ROBUSTA_AI_MODEL_NAME = "Robusta"
 
 
@@ -50,6 +59,7 @@ class TokenCountMetadata(BaseModel):
     system_tokens: int
     user_tokens: int
     tools_to_call_tokens: int
+    assistant_tokens: int
     other_tokens: int
 
 
@@ -90,6 +100,19 @@ class LLM:
     def get_maximum_output_token(self) -> int:
         pass
 
+    def get_max_token_count_for_single_tool(self) -> int:
+        if (
+            0 < TOOL_MAX_ALLOCATED_CONTEXT_WINDOW_PCT
+            and TOOL_MAX_ALLOCATED_CONTEXT_WINDOW_PCT <= 100
+        ):
+            context_window_size = self.get_context_window_size()
+            calculated_max_tokens = int(
+                context_window_size * TOOL_MAX_ALLOCATED_CONTEXT_WINDOW_PCT // 100
+            )
+            return min(calculated_max_tokens, TOOL_MAX_ALLOCATED_CONTEXT_WINDOW_TOKENS)
+        else:
+            return TOOL_MAX_ALLOCATED_CONTEXT_WINDOW_TOKENS
+
     @abstractmethod
     def count_tokens(
         self, messages: list[dict], tools: Optional[list[dict[str, Any]]] = None
@@ -279,6 +302,7 @@ class DefaultLLM(LLM):
         total_tokens = 0
         tools_tokens = 0
         system_tokens = 0
+        assistant_tokens = 0
         user_tokens = 0
         other_tokens = 0
         tools_to_call_tokens = 0
@@ -296,6 +320,8 @@ class DefaultLLM(LLM):
                 user_tokens += token_count
             elif role == "tool":
                 tools_tokens += token_count
+            elif role == "assistant":
+                assistant_tokens += token_count
             else:
                 # although this should not be needed,
                 # it is defensive code so that all tokens are accounted for
@@ -320,6 +346,7 @@ class DefaultLLM(LLM):
             tools_tokens=tools_tokens,
             tools_to_call_tokens=tools_to_call_tokens,
             other_tokens=other_tokens,
+            assistant_tokens=assistant_tokens,
         )
 
     def get_litellm_corrected_name_for_robusta_ai(self) -> str:
@@ -544,6 +571,7 @@ class LLMModelRegistry:
                 return
 
             account_id, token = self.dal.get_ai_credentials()
+
             robusta_models: RobustaModelsResponse | None = fetch_robusta_models(
                 account_id, token
             )

holmes/core/supabase_dal.py

@@ -237,70 +237,61 @@ class SupabaseDal:
             logging.exception("Supabase error while retrieving efficiency data")
             return None
 
-    def get_configuration_changes(
-        self, start_datetime: str, end_datetime: str
+    def get_configuration_changes_metadata(
+        self,
+        start_datetime: str,
+        end_datetime: str,
+        limit: int = 100,
+        workload: Optional[str] = None,
+        ns: Optional[str] = None,
     ) -> Optional[List[Dict]]:
         if not self.enabled:
             return []
 
         try:
-            changes_response = (
+            query = (
                 self.client.table(ISSUES_TABLE)
-                .select("id", "subject_name", "subject_namespace", "description")
+                .select(
+                    "id",
+                    "title",
+                    "subject_name",
+                    "subject_namespace",
+                    "subject_type",
+                    "description",
+                    "starts_at",
+                    "ends_at",
+                )
                 .eq("account_id", self.account_id)
                 .eq("cluster", self.cluster)
                 .eq("finding_type", "configuration_change")
                 .gte("creation_date", start_datetime)
                 .lte("creation_date", end_datetime)
-                .execute()
+                .limit(limit)
             )
-            if not len(changes_response.data):
-                return None
 
-        except Exception:
-            logging.exception("Supabase error while retrieving change data")
-            return None
+            if workload:
+                query.eq("subject_name", workload)
+            if ns:
+                query.eq("subject_namespace", ns)
 
-        changes_ids = [change["id"] for change in changes_response.data]
-        try:
-            change_data_response = (
-                self.client.table(EVIDENCE_TABLE)
-                .select("*")
-                .eq("account_id", self.account_id)
-                .in_("issue_id", changes_ids)
-                .not_.in_("enrichment_type", ENRICHMENT_BLACKLIST)
-                .execute()
-            )
-            if not len(change_data_response.data):
+            res = query.execute()
+            if not res.data:
                 return None
 
-            truncate_evidences_entities_if_necessary(change_data_response.data)
-
         except Exception:
-            logging.exception("Supabase error while retrieving change content")
+            logging.exception("Supabase error while retrieving change data")
             return None
 
-        changes_data = []
-        change_data_map = {
-            change["issue_id"]: change for change in change_data_response.data
-        }
-
-        for change in changes_response.data:
-            change_content = change_data_map.get(change["id"])
-            if change_content:
-                changes_data.append(
-                    {
-                        "change": change_content["data"],
-                        "evidence_id": change_content["id"],
-                        **change,
-                    }
-                )
-
         logging.debug(
-            "Change history for %s-%s: %s", start_datetime, end_datetime, changes_data
+            "Change history metadata for %s-%s workload %s in ns %s: %s",
+            start_datetime,
+            end_datetime,
+            workload,
+            ns,
+            res.data,
         )
 
-        return changes_data
+        return res.data
 
     def unzip_evidence_file(self, data):
         try: