The desync regression test was created as a regression test for the
following bug:
in case we embed NULL termination inside inline/multi-bulk message we
will not be able to perform strchr in order to
identify the newline(\n)/carriage-return(\r) in the client query buffer.
this can influence (for example) replica reading primary stream and keep
filling it's query buffer endlessly consuming more and more memory.
In order to handle the above risk, a check was added to verify the
inline bulk and multi-bulk size are not exceeding the 64K bytes in the
query-buffer. A test was placed in order to verify this.
This PR introduce the following fixes to the desync regression test:
1. fix the sent payload to flush 1024 bytes block of 'A's instead of
'payload' which was sent by mistake.
2. Make sure that the connection is correctly terminated on protocol
error by the server after exceeding the 64K and not over 64K.
3. add another test intrinsic which will also verify the nested bulk
with embedded null termination (was not verified before)
fixes https://github.com/valkey-io/valkey/issues/1583
NOTE: Although it is possible to change the use of strchr to a more
"safe" utility (eg memchr) which will not pause scan at first occurrence
of '\0', we still like to protect against over excessive usage of the
query buffer and also preserve the current behavior(?). We will look
into improving this though in a followup issue.
---------
Signed-off-by: Ran Shidlansik <ranshid@amazon.com>
Signed-off-by: ranshid <88133677+ranshid@users.noreply.github.com>
dd92d079dc