Repository navigation
bug: ReadRows kills ydb-local #15701
Description
To fix this issue, it is necessary to change the current behavior of ReadRows.
The root cause is that ReadRows does not properly handle nullable columns: nullable types in table schema become not-nullable in ReadRows response. That causes ydbd crashes in case of nulls in table data.
repro
CREATE TABLE issues (
id UUID NOT NULL,
title Text,
created_at Timestamp,
author Text,
PRIMARY KEY (id)
);
Sending proto:
session_id: "ydb://session/3?node_id=1&id=OGYyOWJlOWYtOWYyY2Q3YWMtY2UwYWVjOTEtZTllOGRlZmY="
path: "/local/issues"
keys {
type {
list_type {
item {
struct_type {
members {
name: "id"
type {
type_id: UUID
}
}
}
}
}
}
value {
items {
items {
low_128: 5401043504824218602
high_128: 8045957451189022117
}
}
}
}
last logs:
2025-03-13T12:59:53.469553Z :FLAT_TX_SCHEMESHARD NOTICE: Publication complete, notify & remove, at schemeshard: 72075186232723360, txId: 281474976710668, subscribers: 1
2025-03-13T12:59:54.069107Z :READ_TABLE_API NOTICE: [1:7481278092023663194:2485] Finish grpc stream, status: 400000
2025-03-13T12:59:54.149576Z :RPC_REQUEST ERROR: TReadRowsRPC ReplyWithError: no keys are found in request's proto
VERIFY failed (2025-03-13T12:59:54.240681Z): AsValue<T>() type size 8 doesn't match TCell size 0
ydb/core/scheme/scheme_tablecell.h:170
AsValue(): requirement sizeof(T) == Size() failed
??+0 (0x912F72C)
??+0 (0x912AF8C)
??+0 (0xC15585C)
??+0 (0x12D7C120)
??+0 (0x12D7B811)
??+0 (0x12D7E1C1)
??+0 (0x12D7B3F9)
??+0 (0x9BF4B36)
??+0 (0x9BF83F2)
??+0 (0x9BF8041)
??+0 (0x9BF8B53)
??+0 (0x91330C8)
??+0 (0x40008EBAC3)
clone+68 (0x400097CA04)
Metadata
Metadata
Assignees
Labels
No labels
Activity
lex007in commentedon Apr 1, 2025
To fix this issue, it is necessary to change the current behavior of ReadRows.
The root cause is that ReadRows does not properly handle nullable columns: nullable types in table schema become not-nullable in ReadRows response. That causes ydbd crashes in case of nulls in table data.
The current behavior of ReadRows is wrong, so I'll change it in related PR.