File size: 3,924 Bytes
02039a4 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 | #!/usr/bin/env python3
"""Read one cell's episode parquet: list it, unpack an episode, verify the hashes.
python3 scripts/read_episodes.py <cell>.parquet # what is in here
python3 scripts/read_episodes.py <cell>.parquet --extract <entry_id> --out DIR
python3 scripts/read_episodes.py <cell>.parquet --verify # recompute every sha256
Needs pyarrow; nothing else beyond the standard library.
Layout: one row per EPISODE. The artefacts live in the `files` column as
``list<struct<name, kind, size_bytes, sha256, bytes>>`` -- `name` is the path the
file had inside the episode directory, so --extract restores the original tree.
`sha256` is taken over the PUBLISHED bytes (after absolute paths and box names
were scrubbed out of the text payloads), so --verify passes on this file as it
stands; it does not check against anything outside the repository.
"""
import argparse, collections, hashlib, os, sys
import pyarrow.parquet as pq
def rows_of(path, columns=None):
for batch in pq.ParquetFile(path).iter_batches(batch_size=1, columns=columns):
for row in batch.to_pylist():
yield row
def listing(path):
meta = pq.ParquetFile(path).metadata
light = ["entry_id", "task_name", "source_model", "condition", "cell_id",
"is_successful", "score", "file_count", "payload_size_bytes"]
rows = list(rows_of(path, columns=light))
print("%s %.1f MB · %d row(s) · %d row group(s)"
% (os.path.basename(path), os.path.getsize(path) / 1e6,
meta.num_rows, meta.num_row_groups))
if rows:
print("cell: %s" % rows[0]["cell_id"])
print()
print("%-58s %-9s %6s %9s" % ("entry_id", "verdict", "files", "payload"))
for r in rows:
print("%-58s %-9s %6d %8.1fM"
% (r["entry_id"],
"-" if r["is_successful"] is None else
("success" if r["is_successful"] >= 1 else "fail"),
r["file_count"], r["payload_size_bytes"] / 1e6))
kinds = collections.Counter()
for r in rows_of(path, columns=["files"]):
for f in r["files"]:
kinds[f["kind"]] += 1
print("\nfile kinds: %s" % ", ".join("%s %d" % (k, n) for k, n in kinds.most_common()))
def extract(path, entry_id, out_dir):
for r in rows_of(path):
if r["entry_id"] != entry_id:
continue
for f in r["files"]:
dst = os.path.join(out_dir, entry_id, f["name"])
os.makedirs(os.path.dirname(dst), exist_ok=True)
with open(dst, "wb") as fh:
fh.write(f["bytes"])
print("%d file(s) -> %s" % (len(r["files"]), os.path.join(out_dir, entry_id)))
return 0
print("no episode with entry_id %r (use the listing to see them)" % entry_id, file=sys.stderr)
return 1
def verify(path):
# A checker that examines NOTHING must not report success: an empty or
# truncated file would otherwise print "0/0 file(s) match" and exit 0.
n = bad = 0
for r in rows_of(path, columns=["entry_id", "files"]):
for f in r["files"]:
n += 1
if hashlib.sha256(f["bytes"]).hexdigest() != f["sha256"]:
bad += 1
print("MISMATCH %s/%s" % (r["entry_id"], f["name"]))
if n == 0:
print("no payload files found in %s -- nothing was verified" % path, file=sys.stderr)
return 2
print("%d/%d file(s) match their stored sha256" % (n - bad, n))
return 1 if bad else 0
def main():
ap = argparse.ArgumentParser()
ap.add_argument("parquet")
ap.add_argument("--extract", metavar="ENTRY_ID")
ap.add_argument("--out", default=".")
ap.add_argument("--verify", action="store_true")
a = ap.parse_args()
if a.extract:
return extract(a.parquet, a.extract, a.out)
if a.verify:
return verify(a.parquet)
listing(a.parquet)
return 0
sys.exit(main())
|