Kim van der Riet created QPID-5483:
--------------------------------------
Summary: [linearstore] Recovery of journal with partly written
record fails with "JERR_JREC_BADRECTAIL: Invalid data record tail" error message
Key: QPID-5483
URL: https://issues.apache.org/jira/browse/QPID-5483
Project: Qpid
Issue Type: Bug
Components: C++ Broker
Reporter: Kim van der Riet
Assignee: Kim van der Riet
When recovering a store journal which contains a partly written record (such
that the header is present, but the tail record is not), the broker terminates
with the message
{color:blue}{{critical Unexpected error: Queue tq1: recoverMessages() failed:
jexception 0x0701 enq_rec::decode() threw JERR_JREC_BADRECTAIL: Invalid data
record tail. (Magic: expected 0x9aacb3ae; found 0x0)}}{color}
Expected behavior: The broker should handle such recoveries by treating the
partly written record as invalid and, if necessary, adding filler records so
that the next write can take place on an softblock boundary.
To reproduce:
1. Start the broker:
{color:blue}{{./qpidd --load-module linearstore.so -m no --auth no
--default-flow-stop-threshold 0 --default-flow-resume-threshold 0
--default-queue-limit 0 --store-dir /home/kpvdr/RedHat --log-enable info+
--truncate yes}}{color}
2. Run a client to create some records in a queue:
{color:blue}{{./qpid-send -a "tq1;{create:always,node:{type:queue,durable:true}
}" --durable true --tx 0 --content-size 1000 --messages 100}}{color}
3. Use a hex editor (eg hexedit) to partly overwrite the last record from the
middle of the data section (containing the 'X' characters through to the end of
the last write (where all the '\0' characters start) so that the record header
and part of the data is still present, but the tail end of the data and record
tail (and any filler records) are overwritten.
4. Restart the broker to recover the store:
{color:blue}{{./qpidd --load-module linearstore.so -m no --auth no
--default-flow-stop-threshold 0 --default-flow-resume-threshold 0
--default-queue-limit 0 --store-dir /home/kpvdr/RedHat --log-enable info+
--truncate no}}{color}
The broker first detects the bad alignment:
{color:blue}
{{2014-01-15 16:38:45 \[Store\] warning Linear Store: Journal "tq1": Bad record
alignment found at fid=0x1 offs=0x139800 (likely journal overwrite boundary);
16 filler record(s) required.}}
{{2014-01-15 16:38:45 \[Store\] notice Linear Store: Journal "tq1": Recover
phase write: Wrote filler record: fid=0x1 offs=0x139800}}
{{...}}
{color}
but terminates with the error message at the top of this description.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
