download: more consistent and exhaustive logging, new DANDI_DEVEL_AGGRESSIVE_RETRY mode, respect (?) Retry-After

DEVELOPMENT.md

@@ -85,6 +85,10 @@ development command line options.
   function will patch `requests` to log the results of calls to
   `requests.utils.super_len()`
 
+- `DANDI_DOWNLOAD_AGGRESSIVE_RETRY` -- When set, would make `download()` retry
+  very aggressively - it would keep trying if at least some bytes are downloaded
+  on each attempt.  Typically is not needed and could be a sign of network issues.
+
 ## Sourcegraph
 
 The [Sourcegraph](https://sourcegraph.com) browser extension can be used to

dandi/cli/command.py

@@ -114,14 +114,20 @@
             lambda r: r.name != "pyout" and not r.name.startswith("pyout.")
         )
     root.addHandler(handler)
+    exts = (
+        "dandischema",
+        "h5py",
+        "hdmf",
+        "pynwb",
+        "requests",
+        "urllib3",
+    )
 
     lgr.info(
-        "dandi v%s, dandischema v%s, hdmf v%s, pynwb v%s, h5py v%s",
+        "python %s, dandi %s, "
+        + ", ".join("%s %s" % (e, get_module_version(e)) for e in sorted(exts)),
+        sys.version.split()[0],
         __version__,
-        get_module_version("dandischema"),
-        get_module_version("hdmf"),
-        get_module_version("pynwb"),
-        get_module_version("h5py"),
         extra={"file_only": True},
     )
     lgr.info("sys.argv = %r", sys.argv, extra={"file_only": True})

dandi/cli/tests/test_service_scripts.py

@@ -11,7 +11,6 @@
 from click.testing import CliRunner
 from dandischema.consts import DANDI_SCHEMA_VERSION
 import pytest
-import vcr
 
 from dandi import __version__
 from dandi.tests.fixtures import SampleDandiset
@@ -76,9 +75,14 @@ def test_update_dandiset_from_doi(
     dandiset_id = new_dandiset.dandiset_id
     repository = new_dandiset.api.instance.gui
     monkeypatch.setenv("DANDI_API_KEY", new_dandiset.api.api_key)
-    if os.environ.get("DANDI_TESTS_NO_VCR", ""):
+    if os.environ.get("DANDI_TESTS_NO_VCR", "") or sys.version_info <= (3, 10):
+        # Older vcrpy has an issue with Python 3.9 and newer urllib2 >= 2
+        # But we require newer urllib2 for more correct operation, and
+        # do still support 3.9.  Remove when 3.9 support is dropped
         ctx = nullcontext()
     else:
+        import vcr
+
         ctx = vcr.use_cassette(
             str(DATA_DIR / "update_dandiset_from_doi" / f"{name}.vcr.yaml"),
             before_record_request=record_only_doi_requests,

dandi/dandiapi.py

@@ -1592,10 +1592,23 @@
             # TODO: apparently we might need retries here as well etc
             # if result.status_code not in (200, 201):
             result.raise_for_status()
+            nbytes, nchunks = 0, 0
             for chunk in result.iter_content(chunk_size=chunk_size):
+                nchunks += 1
                 if chunk:  # could be some "keep alive"?
+                    nbytes += len(chunk)
                     yield chunk
-            lgr.info("Asset %s successfully downloaded", self.identifier)
+                else:
+                    lgr.debug("'Empty' chunk downloaded for %s", url)
+            lgr.info(
+                "Asset %s (%d bytes in %d chunks starting from %d) successfully "
+                "downloaded from %s",
+                self.identifier,
+                nbytes,
+                nchunks,
+                start_at,
+                url,
+            )
 
         return downloader
 

dandi/download.py

@@ -13,6 +13,7 @@
 from pathlib import Path
 import random
 from shutil import rmtree
+import sys
 from threading import Lock
 import time
 from types import TracebackType
@@ -692,17 +693,26 @@
     # TODO: how do we discover the total size????
     # TODO: do not do it in-place, but rather into some "hidden" file
     resuming = False
-    for attempt in range(3):
+    attempt = 0
+    nattempts = 3  # number to do, could be incremented if we downloaded a little
+    while attempt <= nattempts:
+        attempt += 1
         try:
             if digester:
                 downloaded_digest = digester()  # start empty
             warned = False
             # I wonder if we could make writing async with downloader
             with DownloadDirectory(path, digests or {}) as dldir:
                 assert dldir.offset is not None
+                downloaded_in_attempt = 0
                 downloaded = dldir.offset
                 resuming = downloaded > 0
                 if size is not None and downloaded == size:
+                    lgr.debug(
+                        "%s - downloaded size matches target size of %d, exiting the loop",
+                        path,
+                        size,
+                    )
                     # Exit early when downloaded == size, as making a Range
                     # request in such a case results in a 416 error from S3.
                     # Problems will result if `size` is None but we've already
@@ -713,6 +723,7 @@
                         assert downloaded_digest is not None
                         downloaded_digest.update(block)
                     downloaded += len(block)
+                    downloaded_in_attempt += len(block)
                     # TODO: yield progress etc
                     out: dict[str, Any] = {"done": downloaded}
                     if size:
@@ -738,30 +749,83 @@
         # Catching RequestException lets us retry on timeout & connection
         # errors (among others) in addition to HTTP status errors.
         except requests.RequestException as exc:
+            sleep_amount = random.random() * 5 * attempt
+            if os.environ.get("DANDI_DOWNLOAD_AGGRESSIVE_RETRY"):
+                # in such a case if we downloaded a little more --
+                # consider it a successful attempt
+                if downloaded_in_attempt > 0:
+                    lgr.debug(
+                        "%s - download failed on attempt #%d: %s, "
+                        "but did download %d bytes, so considering "
+                        "it a success and incrementing number of allowed attempts.",
+                        path,
+                        attempt,
+                        exc,
+                        downloaded_in_attempt,
+                    )
+                    nattempts += 1
             # TODO: actually we should probably retry only on selected codes,
-            # and also respect Retry-After
-            if attempt >= 2 or (
-                exc.response is not None
-                and exc.response.status_code
-                not in (
+            if exc.response is not None:
+                if exc.response.status_code not in (
                     400,  # Bad Request, but happened with gider:
                     # https://github.com/dandi/dandi-cli/issues/87
                     *RETRY_STATUSES,
+                ):
+                    lgr.debug(
+                        "%s - download failed due to response %d: %s",
+                        path,
+                        exc.response.status_code,
+                        exc,
+                    )
+                    yield {"status": "error", "message": str(exc)}
+                    return
+                elif retry_after := exc.response.headers.get("Retry-After"):
+                    # playing safe
+                    if not str(retry_after).isdigit():
+                        # our code is wrong, do not crash but issue warning so
+                        # we might get report/fix it up
+                        lgr.warning(
+                            "%s - download failed due to response %d with non-integer"
+                            " Retry-After=%r: %s",
+                            path,
+                            exc.response.status_code,
+                            retry_after,
+                            exc,
+                        )
+                        yield {"status": "error", "message": str(exc)}
+                        return
+                    sleep_amount = int(retry_after)
+                    lgr.debug(
+                        "%s - download failed due to response %d with "
+                        "Retry-After=%d: %s, will sleep and retry",
+                        path,
+                        exc.response.status_code,
+                        sleep_amount,
+                        exc,
+                    )
+                else:
+                    lgr.debug("%s - download failed: %s", path, exc)
+                    yield {"status": "error", "message": str(exc)}
+                    return
+            elif attempt >= nattempts:
+                lgr.debug(
+                    "%s - download failed after %d attempts: %s", path, attempt, exc
                 )
-            ):
-                lgr.debug("%s - download failed: %s", path, exc)
                 yield {"status": "error", "message": str(exc)}
                 return
             # if is_access_denied(exc) or attempt >= 2:
             #     raise
             # sleep a little and retry
-            lgr.debug(
-                "%s - failed to download on attempt #%d: %s, will sleep a bit and retry",
-                path,
-                attempt,
-                exc,
-            )
-            time.sleep(random.random() * 5)
+            else:
+                lgr.debug(
+                    "%s - download failed on attempt #%d: %s, will sleep a bit and retry",
+                    path,
+                    attempt,
+                    exc,
+                )
+            time.sleep(sleep_amount)
+    else:
+        lgr.warning("downloader logic: We should not be here!")
 
     if downloaded_digest and not resuming:
         assert downloaded_digest is not None
@@ -829,16 +893,22 @@
         ):
             # Pick up where we left off, writing to the end of the file
             lgr.debug(
-                "Download directory exists and has matching checksum; resuming download"
+                "%s - download directory exists and has matching checksum(s) %s; resuming download",
+                self.dirpath,
+                matching_algs,
             )
             self.fp = self.writefile.open("ab")
         else:
             # Delete the file (if it even exists) and start anew
             if not chkpath.exists():
-                lgr.debug("Starting new download in new download directory")
+                lgr.debug(
+                    "%s - no prior digests found; starting new download", self.dirpath
+                )
             else:
                 lgr.debug(
-                    "Download directory found, but digests do not match; starting new download"
+                    "%s - download directory found, but digests do not match;"
+                    " starting new download",
+                    self.dirpath,
                 )
             try:
                 self.writefile.unlink()
@@ -857,12 +927,35 @@
         exc_tb: TracebackType | None,
     ) -> None:
         assert self.fp is not None
+        if exc_type is not None or exc_val is not None or exc_tb is not None:
+            lgr.debug(
+                "%s - entered __exit__ with position %d with exception: %s, %s",
+                self.dirpath,
+                self.fp.tell(),
+                exc_type,
+                exc_val,
+            )
+        else:
+            lgr.debug(
+                "%s - entered __exit__ with position %d without any exception",
+                self.dirpath,
+                self.fp.tell(),
+            )
         self.fp.close()
         try:
             if exc_type is None:
                 try:
                     self.writefile.replace(self.filepath)
-                except IsADirectoryError:
+                except (IsADirectoryError, PermissionError) as exc:
+                    if isinstance(exc, PermissionError):
+                        if not (
+                            sys.platform.startswith("win") and self.filepath.is_dir()
+                        ):
+                            raise
+                    lgr.debug(
+                        "Destination path %s is a directory; removing it and retrying",
+                        self.filepath,
+                    )
                     rmtree(self.filepath)
                     self.writefile.replace(self.filepath)
         finally:

dandi/tests/test_download.py

@@ -1,7 +1,10 @@
 from __future__ import annotations
 
 from collections.abc import Callable
+from glob import glob
 import json
+import logging
+from multiprocessing import Manager, Process
 import os
 import os.path
 from pathlib import Path
@@ -21,6 +24,7 @@
 from ..consts import DRAFT, dandiset_metadata_file
 from ..dandiarchive import DandisetURL
 from ..download import (
+    DownloadDirectory,
     Downloader,
     DownloadExisting,
     DownloadFormat,
@@ -1038,3 +1042,89 @@
             assert len(done) == 2
         else:
             assert done[-1] == f"ETA: {10 - i} seconds<"
+
+
+@mark.skipif_on_windows  # https://github.com/pytest-dev/pytest/issues/12964
+def test_DownloadDirectory_basic(tmp_path: Path) -> None:
+    with DownloadDirectory(tmp_path, digests={}) as dl:
+        assert dl.dirpath.exists()
+        assert dl.writefile.exists()
+        assert dl.writefile.stat().st_size == 0
+        assert dl.offset == 0
+
+        dl.append(b"123")
+        assert dl.fp is not None
+        dl.fp.flush()  # appends are not flushed automatically
+        assert dl.writefile.stat().st_size == 3
+        assert dl.offset == 0  # doesn't change
+
+        dl.append(b"456")
+        inode_number = dl.writefile.stat().st_ino
+        assert inode_number != tmp_path.stat().st_ino
+
+    # but after we are done - should be a full file!
+    assert tmp_path.stat().st_size == 6
+    assert tmp_path.read_bytes() == b"123456"
+    # we moved the file, didn't copy (expensive)
+    assert inode_number == tmp_path.stat().st_ino
+
+    # no problem with overwriting with new content
+    with DownloadDirectory(tmp_path, digests={}) as dl:
+        dl.append(b"789")
+    assert tmp_path.read_bytes() == b"789"
+
+    # even if path is a directory which we "overwrite"
+    tmp_path.unlink()
+    tmp_path.mkdir()
+    (tmp_path / "somedata.dat").write_text("content")
+    with DownloadDirectory(tmp_path, digests={}) as dl:
+        assert set(glob(f"{tmp_path}*")) == {str(tmp_path), str(dl.dirpath)}
+        dl.append(b"123")
+    assert tmp_path.read_bytes() == b"123"
+
+    # no temp .dandidownload folder is left behind
+    assert set(glob(f"{tmp_path}*")) == {str(tmp_path)}
+
+    # test locking
+    with Manager() as manager:
+        results = manager.list()
+        with DownloadDirectory(tmp_path, digests={}) as dl:
+            dl.append(b"123")
+            p1 = Process(target=_download_directory_subproc, args=(tmp_path, results))
+            p1.start()
+            p1.join()
+        assert len(results) == 1
+        assert results[0] == f"Could not acquire download lock for {tmp_path}"
+    assert tmp_path.read_bytes() == b"123"
+
+
+# needs to be a top-level function for pickling
+def _download_directory_subproc(path, results):
+    try:
+        with DownloadDirectory(path, digests={}):
+            results.append("re-entered fine")
+    except Exception as exc:
+        results.append(str(exc))
+
+
+def test_DownloadDirectory_exc(
+    tmp_path: Path, caplog: pytest.LogCaptureFixture
+) -> None:
+    caplog.set_level(logging.DEBUG, logger="dandi")
+    # and now let's exit with exception
+    with pytest.raises(RuntimeError):
+        with DownloadDirectory(tmp_path, digests={}) as dl:
+            dl.append(b"456")
+            raise RuntimeError("Boom")
+    assert (
+        "dandi",
+        10,
+        f"{dl.dirpath} - entered __exit__ with position 3 with exception: "
+        "<class 'RuntimeError'>, Boom",
+    ) == caplog.record_tuples[-1]
+    # and we left without cleanup but closed things up after ourselves
+    assert tmp_path.exists()
+    assert tmp_path.is_dir()
+    assert dl.dirpath.exists()
+    assert dl.fp is None
+    assert dl.writefile.read_bytes() == b"456"