Try to fix the non-persisted rocksdb

[Kailai-Wang]

Context

See the linked issue for full context.

Normally we just write once in rpc handler, so set_sync is used.

Maybe there're some improvement room, like only flush it once when necessary? (e.g. program exit? or any hook from gramine)

Track of I/O

before:

2025-04-16T10:34:42.657347Z [INFO][tokio-runtime-worker][rpc_server::methods::pumpx::get_next_intent_id]: intent_id is 0
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- return from write(...) = 0x79
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- clock_gettime(0, 0x52c9230) = 0x0
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- write(2, 0x3491a10, 0x64) ...
2025-04-16T10:34:42.657384Z [INFO][tokio-runtime-worker][intent_core]: Before updating intent_id: 5
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- return from write(...) = 0x64
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- write(13, 0x5380000, 0x5a) ...
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- return from write(...) = 0x5a
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- time(0) = 0x67ff87c2
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- clock_gettime(0, 0x52c9230) = 0x0
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- write(2, 0x3491a10, 0x63) ...
2025-04-16T10:34:42.657505Z [INFO][tokio-runtime-worker][intent_core]: After updating intent_id: 5
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- return from write(...) = 0x63

with this workaround:

2025-04-16T12:57:01.178136Z [DEBUG][tokio-runtime-worker][executor_storage]: before inserting key
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- return from write(...) = 0x62
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- write(13, 0x9cda000, 0x5a) ...
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- return from write(...) = 0x5a
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- fdatasync(13) = 0x0
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- fsync(10) = 0x0
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- time(0) = 0x67ffa91d
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- clock_gettime(0, 0x9c231f0) = 0x0
(libos_parser.c:1701:buf_write_all) [P1:T2:omni-executor] trace: ---- write(2, 0x7de6210, 0x61) ...
2025-04-16T12:57:01.199020Z [DEBUG][tokio-runtime-worker][executor_storage]: after inserting key

[linear]

P-1469 Rocksdb corrupted/not persisted on oe-staging

[BillyWooo]

So opts.set_sync(true); does the trick 👍

[Kailai-Wang]

Hold on - still need a bit fix