jssmith / ssqlite Goto Github PK

the frequency seems to have dropped by avoid extend buffer, but the math seems right. mcheck() catches it sometimes before free/malloc, but not always

merge branch, doesn't recover after timeout

After opening a database some of the basic shell commands are broken, e.g.:

SQLite version 3.21.0 2017-10-24 18:55:49
Enter ".help" for usage hints.
Connected to a transient in-memory database.
Use ".open FILENAME" to reopen on a persistent database.
sqlite> .load ./nfs4
sqlite> .open 192.168.1.57/mytestdb
sqlite> .tables
Error: near "ORDER": syntax error
sqlite>

implemented renaming file in a current directory, but no provision for hard relinking a file in a new directory

read 192.168.1.57/tpcc-nfs-journal offset:0 bytes:1 nfs rpc error illegal filehandle

This appears to occur on the second invocation of the Lambda, following one which terminates abnormally.

this is likely a server issue, but on linux I created a directory owned by uid 0, and
presented credentials as 65534 - I guess the anonymous user

when trying to create a file in the directory, the server rejected the operation because of
permissions, but I was able to remove the (empty) directory.

Running the benchmark command

vfsbench nfs4 192.168.1.57/test-1gb

Leads to this error once the block size reaches 16k.

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff6a92246 in read_beu32 (b=0x1b93000) at ./buffer.h:101
101	    u32 v = ntohl(*(u32*)(b->contents + b->start));

Test data file was created with

head -c 1073741824 /dev/urandom > /efs/test-1gb

if you leave a sqlite command line open for a few tens of minutes and come back to it, the session has timed out, and the client should re-establish synch with the server. it doesn't.

(actually, there is supposed to be a thread/timer sending a SEQUENCE message in some interval defined someplace to keep it alive)

read_command executes the first argument as a command.

This is likely supposed to call the open_command to return the nfs4_file of the argument.

When running TPC-C in Lambda error messages sometimes appear that are not seen when running in EC2, e.g.:

read 192.168.1.57/tpcc-nfs offset:5836800 bytes:4096 status server socket read error

this leads to SQLite failed transactions such as:

[WARNING]	2017-12-18T20:00:06.320Z	0afde782-e42e-11e7-b689-e77e850ec1ca	Failed to execute Transaction 'NEW_ORDER': SQL logic error

optimizations we are consider are

1. pipeline segmented transfers
2. connection bonding
3. iovecs to reduce copying
4. caching filehandles

note that (2) might obviate (1). (1) may in fact be necessary for correctness if the server evaluates out of order

discard_string attempts to discard a string of length 128 which goes beyond the bounds of the buffer b.

discard_string has been modified to skip to the end of the buffer rather than go beyond the bounds of the buffer.

If the program attempts to read again, after a too large discard_string, we'd likely see an "out of data" error.

The structure of can be found in the RFC 7863. I've started laying out the size of the struct in this spreadsheet.

not sure how this happens, and don't have a clear repro, but after screwing around
with .tables and .database and doing some other stuff, with tracing turned on after a while
I get: "status unexpected seq. ID in req"

which (likely?) means that we started to encode something and gave up..or, that
there is some path that doesn't update this..or possibly that there is some specified nfs4
interaction which we aren't managing properly.

awrite tanko generate 5980
Bad session
[5980]
0.003581002
awrite tanko generate 5980
error: encoding mismatch expected 1 got 35 at ../rpc.c:91
0.000109002
awrite tanko generate 5980
[5980]
0.000217002