forked from TrueCloudLab/frostfs-testcases
136 lines
4.6 KiB
Python
136 lines
4.6 KiB
Python
"""
|
|
When doing requests to NeoFS, we get JSON output as an automatically decoded
|
|
structure from protobuf. Some fields are decoded with boilerplates and binary
|
|
values are Base64-encoded.
|
|
|
|
This module contains functions which rearrange the structure and reencode binary
|
|
data from Base64 to Base58.
|
|
"""
|
|
|
|
import base64
|
|
|
|
import base58
|
|
|
|
|
|
def decode_simple_header(data: dict) -> dict:
|
|
"""
|
|
This function reencodes Simple Object header and its attributes.
|
|
"""
|
|
try:
|
|
data = decode_common_fields(data)
|
|
|
|
# Normalize object attributes
|
|
data["header"]["attributes"] = {
|
|
attr["key"]: attr["value"] for attr in data["header"]["attributes"]
|
|
}
|
|
except Exception as exc:
|
|
raise ValueError(f"failed to decode JSON output: {exc}") from exc
|
|
|
|
return data
|
|
|
|
|
|
def decode_split_header(data: dict) -> dict:
|
|
"""
|
|
This function rearranges Complex Object header.
|
|
The header holds SplitID, a random unique
|
|
number, which is common among all splitted objects, and IDs of the Linking
|
|
Object and the last splitted Object.
|
|
"""
|
|
try:
|
|
data["splitId"] = json_reencode(data["splitId"])
|
|
data["lastPart"] = json_reencode(data["lastPart"]["value"]) if data["lastPart"] else None
|
|
data["link"] = json_reencode(data["link"]["value"]) if data["link"] else None
|
|
except Exception as exc:
|
|
raise ValueError(f"failed to decode JSON output: {exc}") from exc
|
|
|
|
return data
|
|
|
|
|
|
def decode_linking_object(data: dict) -> dict:
|
|
"""
|
|
This function reencodes Linking Object header.
|
|
It contains IDs of child Objects and Split Chain data.
|
|
"""
|
|
try:
|
|
data = decode_simple_header(data)
|
|
split = data["header"]["split"]
|
|
split["children"] = [json_reencode(item["value"]) for item in split["children"]]
|
|
split["splitID"] = json_reencode(split["splitID"])
|
|
split["previous"] = json_reencode(split["previous"]["value"]) if split["previous"] else None
|
|
split["parent"] = json_reencode(split["parent"]["value"]) if split["parent"] else None
|
|
except Exception as exc:
|
|
raise ValueError(f"failed to decode JSON output: {exc}") from exc
|
|
|
|
return data
|
|
|
|
|
|
def decode_storage_group(data: dict) -> dict:
|
|
"""
|
|
This function reencodes Storage Group header.
|
|
"""
|
|
try:
|
|
data = decode_common_fields(data)
|
|
except Exception as exc:
|
|
raise ValueError(f"failed to decode JSON output: {exc}") from exc
|
|
|
|
return data
|
|
|
|
|
|
def decode_tombstone(data: dict) -> dict:
|
|
"""
|
|
This function reencodes Tombstone header.
|
|
"""
|
|
try:
|
|
data = decode_simple_header(data)
|
|
data["header"]["sessionToken"] = decode_session_token(data["header"]["sessionToken"])
|
|
except Exception as exc:
|
|
raise ValueError(f"failed to decode JSON output: {exc}") from exc
|
|
return data
|
|
|
|
|
|
def decode_session_token(data: dict) -> dict:
|
|
"""
|
|
This function reencodes a fragment of header which contains
|
|
information about session token.
|
|
"""
|
|
target = data["body"]["object"]["target"]
|
|
target["container"] = json_reencode(target["container"]["value"])
|
|
target["objects"] = [json_reencode(obj["value"]) for obj in target["objects"]]
|
|
return data
|
|
|
|
|
|
def json_reencode(data: str) -> str:
|
|
"""
|
|
According to JSON protocol, binary data (Object/Container/Storage Group IDs, etc)
|
|
is converted to string via Base58 encoder. But we usually operate with Base64-encoded format.
|
|
This function reencodes given Base58 string into the Base64 one.
|
|
"""
|
|
return base58.b58encode(base64.b64decode(data)).decode("utf-8")
|
|
|
|
|
|
def encode_for_json(data: str) -> str:
|
|
"""
|
|
This function encodes binary data for sending them as protobuf
|
|
structures.
|
|
"""
|
|
return base64.b64encode(base58.b58decode(data)).decode("utf-8")
|
|
|
|
|
|
def decode_common_fields(data: dict) -> dict:
|
|
"""
|
|
Despite of type (simple/complex Object, Storage Group, etc) every Object
|
|
header contains several common fields.
|
|
This function rearranges these fields.
|
|
"""
|
|
data["objectID"] = json_reencode(data["objectID"]["value"])
|
|
|
|
header = data["header"]
|
|
header["containerID"] = json_reencode(header["containerID"]["value"])
|
|
header["ownerID"] = json_reencode(header["ownerID"]["value"])
|
|
header["payloadHash"] = json_reencode(header["payloadHash"]["sum"])
|
|
header["version"] = f"{header['version']['major']}{header['version']['minor']}"
|
|
# Homomorphic hash is optional and its calculation might be disabled in trusted network
|
|
if header.get("homomorphicHash"):
|
|
header["homomorphicHash"] = json_reencode(header["homomorphicHash"]["sum"])
|
|
|
|
return data
|