anyscale 0.26.32__py3-none-any.whl → 0.26.34__py3-none-any.whl

anyscale/commands/list_util.py

@@ -32,7 +32,47 @@ def _paginate(iterator: Iterator[Any], page_size: Optional[int]) -> Iterator[Lis
             yield page
 
 
-def display_list(  # noqa: PLR0913
+def _render_page(
+    page: List[Any],
+    item_formatter: Callable[[Any], Dict[str, Any]],
+    table_creator: Callable[[bool], Table],
+    json_output: bool,
+    is_first: bool,
+    page_num: int,
+    console: Console,
+) -> int:
+    """Render a single page of items."""
+    if page_num > 1:  # Only show page number for pages after first
+        console.print(f"[dim]Page {page_num}[/dim]")
+
+    rows = [item_formatter(item) for item in page]
+    if json_output:
+        json_str = json_dumps(rows, indent=2, cls=AnyscaleJSONEncoder)
+        console.print_json(json=json_str)
+    else:
+        tbl = table_creator(is_first)
+        for row in rows:
+            tbl.add_row(*row.values())
+        console.print(tbl)
+
+    return len(page)
+
+
+def _should_continue_pagination(
+    page_size: int, current_page_size: int, console: Console
+) -> bool:
+    """Prompt user to continue pagination if needed."""
+    if current_page_size < page_size:
+        return False  # Last page, no need to prompt
+
+    console.print()
+    console.print(
+        "[dim]Press [bold]Enter[/bold] to continue, [bold]q[/bold] to quit…[/]"
+    )
+    return input("> ").strip().lower() != "q"
+
+
+def display_list(  # noqa: PLR0913, PLR0912
     iterator: Iterator[Any],
     item_formatter: Callable[[Any], Dict[str, Any]],
     table_creator: Callable[[bool], Table],
@@ -65,56 +105,78 @@ def display_list(  # noqa: PLR0913
     total_count = 0
     pages = _paginate(iterator, page_size if interactive else max_items)
 
-    # fetch first page under spinner
+    # Start interactive session if needed
+    if interactive:
+        try:
+            from anyscale.telemetry import start_interactive_session
+
+            start_interactive_session()
+        except Exception:  # noqa: BLE001
+            pass
+
+    # Fetch and render first page
     with console.status("Retrieving items…", spinner="dots"):
         try:
             first_page = next(pages)
         except StopIteration:
             first_page = []
 
-    def _render(page: List[Any], is_first: bool, page_num: int):
-        nonlocal total_count
-        total_count += len(page)
-        if interactive:
-            console.print(f"[dim]Page {page_num}[/dim]")
-        rows = [item_formatter(item) for item in page]
-        if json_output:
-            json_str = json_dumps(rows, indent=2, cls=AnyscaleJSONEncoder)
-            console.print_json(json=json_str)
-        else:
-            tbl = table_creator(is_first)
-            for row in rows:
-                tbl.add_row(*row.values())
-            console.print(tbl)
-
-    # render first page
     if first_page:
-        _render(first_page, True, page_num=1)
+        total_count += _render_page(
+            first_page, item_formatter, table_creator, json_output, True, 1, console
+        )
 
-    # non-interactive: stop after first page
+    # For interactive commands, mark when command logic completes
+    if interactive:
+        try:
+            from anyscale.telemetry import mark_command_complete
+
+            mark_command_complete()
+        except Exception:  # noqa: BLE001
+            pass
+
+    # Non-interactive: stop after first page
     if not interactive:
         return total_count
 
-    # interactive: prompt after full first page
-    if len(first_page) == page_size:
-        console.print()
-        console.print(
-            "[dim]Press [bold]Enter[/bold] to continue, [bold]q[/bold] to quit…[/]"
-        )
-        if input("> ").strip().lower() == "q":
-            return total_count
+    # Interactive: check if user wants to continue
+    if not _should_continue_pagination(page_size, len(first_page), console):
+        return total_count
 
-    # render remaining pages
+    # Render remaining pages with correct telemetry timing
     page_num = 2
-    for page in pages:
-        _render(page, False, page_num)
-        if len(page) == page_size:
-            console.print()
-            console.print(
-                "[dim]Press [bold]Enter[/bold] to continue, [bold]q[/bold] to quit…[/]"
-            )
-            if input("> ").strip().lower() == "q":
-                break
+    while True:
+        # Start page fetch timing and generate new trace ID BEFORE fetching
+        try:
+            from anyscale.telemetry import mark_page_fetch_start
+
+            mark_page_fetch_start(page_num)
+        except Exception:  # noqa: BLE001
+            pass
+
+        # Now fetch the page (with the new trace ID)
+        try:
+            page = next(pages)
+        except StopIteration:
+            break
+
+        # Render the page
+        total_count += _render_page(
+            page, item_formatter, table_creator, json_output, False, page_num, console
+        )
+
+        # Complete page fetch telemetry
+        try:
+            from anyscale.telemetry import mark_page_fetch_complete
+
+            mark_page_fetch_complete(page_num)
+        except Exception:  # noqa: BLE001
+            pass
+
+        # Check if user wants to continue or if this was the last page
+        if not _should_continue_pagination(page_size, len(page), console):
+            break
+
         page_num += 1
 
     return total_count

anyscale/integrations.py

@@ -20,8 +20,6 @@ WANDB_API_KEY_NAME = "WANDB_API_KEY_NAME"  # pragma: allowlist secret
 WANDB_PROJECT_NAME = "WANDB_PROJECT_NAME"
 WANDB_GROUP_NAME = "WANDB_GROUP_NAME"
 
-FLAG_WANDB_INTEGRATION_PROTOTYPE = "wandb-integration-prototype"
-
 log = BlockLogger()  # Anyscale CLI Logger
 
 
@@ -131,13 +129,6 @@ def wandb_setup_api_key_hook() -> Optional[str]:
     be called by the OSS WandbLoggerCallback. Because this is called
     before wandb.init(), any other setup can also be done here.
     """
-    api_client = get_auth_api_client(log_output=False).api_client
-    feature_flag_on = api_client.check_is_feature_flag_on_api_v2_userinfo_check_is_feature_flag_on_get(
-        FLAG_WANDB_INTEGRATION_PROTOTYPE
-    ).result.is_on
-    if not feature_flag_on:
-        return None
-
     protected_api_key = wandb_get_api_key()
 
     try:
@@ -171,11 +162,6 @@ def set_wandb_project_group_env_vars():
     for production jobs, workspaces, and Ray jobs.
     """
     api_client = get_auth_api_client(log_output=False).api_client
-    feature_flag_on = api_client.check_is_feature_flag_on_api_v2_userinfo_check_is_feature_flag_on_get(
-        FLAG_WANDB_INTEGRATION_PROTOTYPE
-    ).result.is_on
-    if not feature_flag_on:
-        return
 
     wandb_project_default = None
     wandb_group_default = None
@@ -224,12 +210,6 @@ def wandb_send_run_info_hook(run: Any) -> None:
     api_client = auth_api_client.api_client
     anyscale_api_client = auth_api_client.anyscale_api_client
 
-    feature_flag_on = api_client.check_is_feature_flag_on_api_v2_userinfo_check_is_feature_flag_on_get(
-        FLAG_WANDB_INTEGRATION_PROTOTYPE
-    ).result.is_on
-    if not feature_flag_on:
-        return
-
     try:
         import wandb
     except ImportError:

anyscale/scripts.py

@@ -43,6 +43,7 @@ from anyscale.commands.user_commands import user_cli
 from anyscale.commands.workspace_commands import workspace_cli
 from anyscale.commands.workspace_commands_v2 import workspace_cli as workspace_cli_v2
 import anyscale.conf
+import anyscale.telemetry  # IMPORTANT: auto-patches click instrumentation on import
 from anyscale.utils.cli_version_check_util import log_warning_if_version_needs_upgrade
 
 

anyscale/shared_anyscale_utils/headers.py

@@ -25,6 +25,10 @@ class RequestHeaders(str, Enum):
     # Identifies the source of the client
     SOURCE = "X-Anyscale-Source"
 
+    # W3C Trace Context traceparent header
+    # https://www.w3.org/TR/trace-context/#traceparent-header
+    TRACEPARENT = "traceparent"
+
 
 class ResponseHeaders(str, Enum):
     """

anyscale/telemetry.py

@@ -0,0 +1,424 @@
+"""
+Telemetry for Anyscale CLI commands.
+
+Patches Click to capture execution metrics for _leaf_ commands,
+including command path, flags, timing, and errors. Emits via
+HTTP POST (best-effort) or debug print.
+
+Supports session-based distributed tracing for interactive commands:
+- Each command gets a unique trace_id for backend correlation
+- Interactive sessions get a session_id to group related operations
+- Page fetches get new trace_ids but share the session_id
+"""
+
+from contextvars import ContextVar
+import functools
+import json
+import os
+import random
+import secrets
+import sys
+import threading
+import time
+from typing import List, Optional
+
+import click
+
+from anyscale.cli_logger import BlockLogger
+from anyscale.client.openapi_client.models.cli_usage_payload import CLIUsagePayload
+
+
+# ─── Configuration ────────────────────────────────────────────────────────────
+
+SAMPLE_RATE = float(os.getenv("ANYSCALE_TELEMETRY_SAMPLE_RATE", "1.0"))
+TELEMETRY_DEBUG = os.getenv("ANYSCALE_DEBUG") == "1"
+
+# ContextVar automatically propagates into asyncio tasks if you ever go async.
+# (Each CLI invocation gets its own interpreter, so this never crosses commands.)
+_trace_id_var: ContextVar[Optional[str]] = ContextVar("_trace_id_var", default=None)
+_session_id_var: ContextVar[Optional[str]] = ContextVar("_session_id_var", default=None)
+_skip_click_patch_var: ContextVar[bool] = ContextVar(
+    "_skip_click_patch_var", default=False
+)
+
+logger = BlockLogger()
+
+# ─── Trace Context Helpers ───────────────────────────────────────────────────
+
+
+def _setup_trace_context() -> str:
+    """Ensure we have a trace ID in the ContextVar, and return it."""
+    try:
+        tid = _trace_id_var.get()
+        if tid is None:
+            tid = secrets.token_hex(16)
+            _trace_id_var.set(tid)
+        logger.debug(f"[TRACE DEBUG] trace-id={tid}")
+        return tid
+    except Exception:  # noqa: BLE001
+        # Fallback to a default trace ID if anything goes wrong
+        return secrets.token_hex(16)
+
+
+def get_traceparent() -> Optional[str]:
+    """Return a W3C-style traceparent header, or None if not initialized."""
+    try:
+        tid = _trace_id_var.get()
+        if not tid:
+            return None
+        return f"00-{tid}-{'0'*16}-01"
+    except Exception:  # noqa: BLE001
+        return None
+
+
+def start_interactive_session() -> str:
+    """Start an interactive session and return the session ID."""
+    try:
+        session_id = secrets.token_hex(8)
+        _session_id_var.set(session_id)
+        logger.debug(f"[TRACE DEBUG] session-id={session_id}")
+        return session_id
+    except Exception:  # noqa: BLE001
+        # Return a fallback session ID
+        return secrets.token_hex(8)
+
+
+def new_trace_for_page() -> str:
+    """Generate a new trace ID for the next page in an interactive session."""
+    try:
+        new_trace_id = secrets.token_hex(16)
+        _trace_id_var.set(new_trace_id)
+        logger.debug(f"[TRACE DEBUG] new-trace-id={new_trace_id}")
+        return new_trace_id
+    except Exception:  # noqa: BLE001
+        # Return a fallback trace ID
+        return secrets.token_hex(16)
+
+
+# ─── CLI Arg Extraction ───────────────────────────────────────────────────────
+
+
+def _get_user_flags() -> List[str]:
+    """Return all `-x`/`--long` flags from the raw argv (no values)."""
+    try:
+        args = sys.argv[1:]
+        # Strip off the program name if Click added it
+        if args and args[0] in ("anyscale", "main"):
+            args = args[1:]
+        return [a for a in args if a.startswith("-")]
+    except Exception:  # noqa: BLE001
+        return []
+
+
+def _get_user_options(ctx: click.Context) -> List[str]:
+    """Return the names of parameters explicitly set via the CLI."""
+    try:
+        opts: List[str] = []
+        for name in ctx.params:
+            try:
+                if (
+                    ctx.get_parameter_source(name)
+                    is click.core.ParameterSource.COMMANDLINE
+                ):
+                    opts.append(name)
+            except Exception:  # noqa: BLE001
+                opts.append(name)
+        return opts
+    except Exception:  # noqa: BLE001
+        return []
+
+
+# ─── Page Fetch Tracking ─────────────────────────────────────────────────────
+
+_page_fetch_start_time: ContextVar[Optional[float]] = ContextVar(
+    "_page_fetch_start_time", default=None
+)
+
+
+def mark_page_fetch_start(page_number: int) -> None:
+    """
+    Mark the start of a page fetch operation. This will:
+    1. Generate a new trace ID for this page
+    2. Start timing the fetch operation
+
+    Args:
+        page_number: The page number being fetched (1-indexed)
+    """
+    try:
+        if SAMPLE_RATE <= 0 or random.random() > SAMPLE_RATE:
+            return
+
+        # Generate new trace ID for this page BEFORE making the API request
+        new_trace_for_page()
+
+        # Start timing
+        _page_fetch_start_time.set(time.perf_counter())
+
+        logger.debug(f"[TRACE DEBUG] page-fetch-start page={page_number}")
+    except Exception:  # noqa: BLE001
+        # Telemetry should never crash the CLI
+        pass
+
+
+def mark_page_fetch_complete(page_number: int) -> None:
+    """
+    Mark the completion of a page fetch operation and emit telemetry.
+    This calculates the duration and sends the page_fetch event.
+
+    Args:
+        page_number: The page number that was fetched (1-indexed)
+    """
+    try:
+        if SAMPLE_RATE <= 0 or random.random() > SAMPLE_RATE:
+            return
+
+        # Calculate duration
+        start_time = _page_fetch_start_time.get()
+        if start_time is None:
+            # Fallback if timing wasn't started properly
+            duration_ms = 0.0
+        else:
+            duration_ms = (time.perf_counter() - start_time) * 1000
+
+        # Get current click context
+        try:
+            ctx = click.get_current_context()
+        except RuntimeError:
+            return  # No active context
+
+        # Get current trace ID (should be the one we generated in mark_page_fetch_start)
+        trace_id = _trace_id_var.get()
+        if not trace_id:
+            return
+
+        # Emit page fetch telemetry
+        body = _create_payload(
+            trace_id=trace_id,
+            ctx=ctx,
+            duration_ms=duration_ms,
+            exit_code=0,
+            exception_type=None,
+            event_type="page_fetch",
+            page_number=page_number,
+        )
+        _emit_telemetry(body)
+
+        # Reset timing
+        _page_fetch_start_time.set(None)
+
+        logger.debug(
+            f"[TRACE DEBUG] page-fetch-complete page={page_number} duration={duration_ms:.2f}ms"
+        )
+    except Exception:  # noqa: BLE001
+        # Telemetry should never crash the CLI
+        pass
+
+
+# ─── Payload Construction ────────────────────────────────────────────────────
+
+
+def _create_payload(
+    trace_id: str,
+    ctx: click.Context,
+    duration_ms: float,
+    exit_code: int,
+    exception_type: Optional[str],
+    event_type: str = "command",
+    page_number: Optional[int] = None,
+) -> CLIUsagePayload:
+    """
+    Build a Typed CLIUsagePayload (from the generated OpenAPI models)
+    so we get IDE/type-checker support on all the fields.
+
+    Args:
+        trace_id: Unique trace identifier for this operation
+        ctx: Click context containing command information
+        duration_ms: Command/operation duration in milliseconds
+        exit_code: Command exit code (0 for success, 1 for error)
+        exception_type: Exception class name if command failed
+        event_type: Type of event ("command" or "page_fetch")
+        page_number: Page number for page_fetch events
+    """
+    try:
+        # Get session ID if available
+        session_id = _session_id_var.get()
+
+        data = {
+            "trace_id": trace_id,
+            "session_id": session_id,
+            "event_type": event_type,
+            "page_number": page_number,
+            "cmd_path": ctx.command_path,
+            "options": sorted(_get_user_options(ctx)),
+            "flags_used": sorted(_get_user_flags()),
+            "duration_ms": round(duration_ms, 2),
+            "exit_code": exit_code,
+            "exception_type": exception_type,
+            "cli_version": getattr(sys.modules.get("anyscale"), "__version__", None),
+            "python_version": f"{sys.version_info.major}.{sys.version_info.minor}",
+            "timestamp": int(time.time()),
+        }
+        return CLIUsagePayload(**data)
+    except Exception:  # noqa: BLE001
+        # Fallback payload with minimal data if construction fails
+        fallback_data = {
+            "trace_id": trace_id,
+            "cmd_path": "unknown",
+            "duration_ms": duration_ms,
+            "exit_code": exit_code,
+            "timestamp": int(time.time()),
+        }
+        return CLIUsagePayload(**fallback_data)
+
+
+def mark_command_complete() -> None:
+    """
+    Mark that the command logic has completed and emit telemetry immediately.
+    For interactive commands, call this when data is ready but before user interaction.
+    This will prevent the Click patch from double-emitting.
+    """
+    try:
+        trace_id = _trace_id_var.get()
+        if not trace_id:
+            return
+
+        # Get current click context
+        try:
+            ctx = click.get_current_context()
+        except RuntimeError:
+            return  # No active context
+
+        # Calculate duration from the click context if available
+        # For interactive commands, we want the time up to this point
+        start_time = getattr(ctx, "telemetry_start_time", None)
+        if start_time is None:
+            # Fallback: use a minimal duration
+            duration_ms = 0.0
+        else:
+            duration_ms = (time.perf_counter() - start_time) * 1000
+
+        # Emit the command completion event
+        body = _create_payload(
+            trace_id=trace_id,
+            ctx=ctx,
+            duration_ms=duration_ms,
+            exit_code=0,
+            exception_type=None,
+            event_type="command",
+        )
+        _emit_telemetry(body)
+
+        # Prevent Click patch from emitting again
+        _skip_click_patch_var.set(True)
+    except Exception:  # noqa: BLE001
+        # Telemetry should never crash the CLI
+        pass
+
+
+# ─── Emission (fire-&-forget) ─────────────────────────────────────────────────
+
+
+def _emit_telemetry(body: CLIUsagePayload) -> None:
+    """
+    Send the payload to the console API. Runs in a short-lived thread
+    so we never block the CLI for more than ~3 seconds.
+    """
+    try:
+        logger.debug(json.dumps(body.to_dict(), indent=2))
+
+        def _worker():
+            try:
+                # Lazy imports to avoid circular deps
+                from anyscale.authenticate import get_auth_api_client
+                from anyscale.client.openapi_client.api.default_api import DefaultApi
+
+                auth_block = get_auth_api_client()
+                api = DefaultApi(api_client=auth_block.anyscale_api_client)
+                api.receive_cli_usage_api_v2_cli_usage_post(
+                    cli_usage_payload=body, _request_timeout=2
+                )
+            except Exception:  # noqa: BLE001
+                # Best-effort only - never crash the CLI
+                pass
+
+        thread = threading.Thread(target=_worker, daemon=False)
+        thread.start()
+        thread.join(timeout=3)
+    except Exception:  # noqa: BLE001
+        # Telemetry should never crash the CLI
+        pass
+
+
+# ─── Click Patch ─────────────────────────────────────────────────────────────
+
+
+def _patch_click() -> None:
+    """Monkey-patch Click so that each leaf command emits telemetry."""
+    try:
+        if getattr(click, "_anyscale_telemetry_patched", False):
+            return
+
+        original_invoke = click.Command.invoke
+
+        @functools.wraps(original_invoke)
+        def instrumented_invoke(self, ctx, *args, **kwargs):
+            try:
+                # Sampling
+                if SAMPLE_RATE <= 0 or random.random() > SAMPLE_RATE:
+                    return original_invoke(self, ctx, *args, **kwargs)
+                # Only instrument leaf commands
+                if isinstance(self, click.Group):
+                    return original_invoke(self, ctx, *args, **kwargs)
+
+                trace_id = _setup_trace_context()
+                start = time.perf_counter()
+
+                # Store start time in context for interactive commands
+                ctx.telemetry_start_time = start
+
+                code, exc = 0, None
+                should_emit_telemetry = True
+
+                try:
+                    result = original_invoke(self, ctx, *args, **kwargs)
+                    return result
+                except Exception as e:
+                    code, exc = 1, e.__class__.__name__
+                    raise
+                finally:
+                    # Only emit telemetry once per command invocation
+                    if _skip_click_patch_var.get():
+                        should_emit_telemetry = False
+
+                    if should_emit_telemetry:
+                        try:
+                            # Use actual end time for non-interactive commands
+                            dur = (time.perf_counter() - start) * 1_000
+                            body = _create_payload(
+                                trace_id=trace_id,
+                                ctx=ctx,
+                                duration_ms=dur,
+                                exit_code=code,
+                                exception_type=exc,
+                                event_type="command",
+                            )
+                            _emit_telemetry(body)
+                            _skip_click_patch_var.set(True)
+                        except Exception:  # noqa: BLE001
+                            # Telemetry should never crash the CLI
+                            pass
+            except Exception:  # noqa: BLE001
+                # If telemetry setup fails, just run the original command
+                return original_invoke(self, ctx, *args, **kwargs)
+
+        click.Command.invoke = instrumented_invoke
+        click._anyscale_telemetry_patched = (  # noqa: SLF001  # type: ignore[attr-defined]
+            True
+        )
+    except Exception:  # noqa: BLE001
+        # If patching fails, telemetry just won't work - don't crash the CLI
+        pass
+
+
+# Auto-patch on import
+_patch_click()

anyscale/version.py

@@ -1 +1 @@
-__version__ = "0.26.32"
+__version__ = "0.26.34"

{anyscale-0.26.32.dist-info → anyscale-0.26.34.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: anyscale
-Version: 0.26.32
+Version: 0.26.34
 Summary: Command Line Interface for Anyscale
 Author: Anyscale Inc.
 License: AS License