Dump state file atomically not to corrupt it

[Al2Klimov]

by using fsync(2) before close(2) and rename(2).

closes #9446

[Al2Klimov]

Test

Start strace -f icinga2 daemon -z and send a term signal to the umbrella process.

[pid  5637] write(1, "[2022-07-21 18:53:03 +0200] \33[32"..., 126[2022-07-21 18:53:03 +0200] information/ConfigObject: Dumping program state to file '/var/lib/icinga2/icinga2.state'
) = 126
[pid  5637] openat(AT_FDCWD, "/var/lib/icinga2", O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_DIRECTORY) = 12
[pid  5637] fstat(12, {st_mode=S_IFDIR|0750, st_size=4096, ...}) = 0
[pid  5637] getdents64(12, 0x5642e03e7580 /* 5 entries */, 32768) = 160
[pid  5637] getdents64(12, 0x5642e03e7580 /* 0 entries */, 32768) = 0
[pid  5637] close(12)                   = 0
[pid  5637] getpid()                    = 5637
[pid  5637] openat(AT_FDCWD, "/var/lib/icinga2/icinga2.state.tmp.wReKTr", O_RDWR|O_CREAT|O_EXCL, 0600) = 12
[pid  5637] openat(AT_FDCWD, "/var/lib/icinga2/icinga2.state.tmp.wReKTr", O_WRONLY|O_CREAT|O_TRUNC, 0666) = 14
[pid  5637] close(12)                   = 0
[pid  5637] chmod("/var/lib/icinga2/icinga2.state.tmp.wReKTr", 0600) = 0
[pid  5637] write(14, "81:{\"name\":\"ido\",\"type\":\"CheckCo"..., 270) = 270
[pid  5637] fsync(14)                   = 0
[pid  5637] close(14)                   = 0
[pid  5637] rename("/var/lib/icinga2/icinga2.state.tmp.wReKTr", "/var/lib/icinga2/icinga2.state") = 0

Awesome! If this doesn’t help... I don’t know.

[Al2Klimov]

TODO

• Build error on Windows
• Test on Windows
• Packaging: build deps
• Re-run GHA

[julianbrost]

Ah, looks like some parts of Boost.Iostreams aren't header-only in particular file_descriptor: https://www.boost.org/doc/libs/1_79_0/libs/iostreams/doc/installation.html

Sounds like a bit of extra fun then, but okay.

[Al2Klimov]

My compromise implementation:

[Al2Klimov]

[pid  5067] write(1, "[2022-07-26 16:44:27 +0200] \33[32"..., 126[2022-07-26 16:44:27 +0200] information/ConfigObject: Dumping program state to file '/var/lib/icinga2/icinga2.state'
) = 126
[pid  5067] openat(AT_FDCWD, "/var/lib/icinga2", O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_DIRECTORY) = 12
[pid  5067] fstat(12, {st_mode=S_IFDIR|0750, st_size=4096, ...}) = 0
[pid  5067] getdents64(12, 0x5608086ab890 /* 5 entries */, 32768) = 160
[pid  5067] getdents64(12, 0x5608086ab890 /* 0 entries */, 32768) = 0
[pid  5067] close(12)                   = 0
[pid  5067] getpid()                    = 5067
[pid  5067] openat(AT_FDCWD, "/var/lib/icinga2/icinga2.state.tmp.bd3lAk", O_RDWR|O_CREAT|O_EXCL, 0600) = 12
[pid  5067] chmod("/var/lib/icinga2/icinga2.state.tmp.bd3lAk", 0600) = 0
[pid  5067] write(12, "81:{\"name\":\"ido\",\"type\":\"CheckCo"..., 270) = 270
[pid  5067] fsync(12)                   = 0
[pid  5067] close(12)                   = 0
[pid  5067] rename("/var/lib/icinga2/icinga2.state.tmp.bd3lAk", "/var/lib/icinga2/icinga2.state") = 0

[Al2Klimov]

In this variant we keep the legacy broken and CreateTempFile() as-is. But we don’t depend on it in new code.