oxedyne/fe2o3/fe2o3_o3db_sync/tests/dist_kill9.rs
7.3 KiB, 1 run
created by r1870400018:67680, which is this file's identity for as long as the history lasts, whatever it is later renamed to
download · who wrote it · its history
| 1 | #![cfg(feature = "dist")] |
| 2 | //! A put the engine returned for survives `kill -9` of the process (A2 QA agent 2). |
| 3 | //! |
| 4 | //! The one-zone configuration that `DistOzone::over_o3db` is used with in production: one zone, |
| 5 | //! a barrier on every write and a cache far smaller than what is stored. The test re-runs its |
| 6 | //! own binary as a child that puts records in a loop and prints each id once `put_at` has |
| 7 | //! returned for it. The parent reads those lines, kills the child with SIGKILL in mid-stream, |
| 8 | //! reopens the root, and every acknowledged record must read back byte for byte. Three rounds |
| 9 | //! reopen a root that the previous round crashed. The one write the child was inside when it |
| 10 | //! died may be there or not, but if it is there it must be whole. |
| 11 | //! |
| 12 | //! Each run takes about half a minute at most. |
| 13 | //! |
| 14 | //! [Written with AI entirely](https://need2know.ai/entirely-ai/code)\ |
| 15 | //! Anthropic Claude |
| 16 | |
| 17 | use oxedyne_fe2o3_core::prelude::*; |
| 18 | use oxedyne_fe2o3_crypto::enc::EncryptionScheme; |
| 19 | use oxedyne_fe2o3_hash::{ |
| 20 | csum::ChecksumScheme, |
| 21 | hash::HashScheme, |
| 22 | }; |
| 23 | use oxedyne_fe2o3_jdat::prelude::*; |
| 24 | use oxedyne_fe2o3_o3db_sync::{ |
| 25 | base::cfg::OzoneConfig, |
| 26 | data::core::RestSchemesInput, |
| 27 | dist::{ |
| 28 | config::{ |
| 29 | DistOzoneConfig, |
| 30 | TableConfig, |
| 31 | }, |
| 32 | engine::DistOzone, |
| 33 | o3db_storage::O3dbStorage, |
| 34 | record::{ |
| 35 | Record, |
| 36 | RecordId, |
| 37 | }, |
| 38 | resolve::LastVersionWins, |
| 39 | storage::Storage, |
| 40 | }, |
| 41 | kademlia::id::NodeId, |
| 42 | oam::config::OamConfig, |
| 43 | test::setup, |
| 44 | }; |
| 45 | |
| 46 | use std::{ |
| 47 | collections::BTreeMap, |
| 48 | io::{ |
| 49 | BufRead, |
| 50 | BufReader, |
| 51 | }, |
| 52 | os::unix::process::ExitStatusExt, |
| 53 | path::PathBuf, |
| 54 | process::{ |
| 55 | Command, |
| 56 | Stdio, |
| 57 | }, |
| 58 | time::Duration, |
| 59 | }; |
| 60 | |
| 61 | |
| 62 | type Eng = DistOzone< |
| 63 | O3dbStorage<{ setup::UID_LEN }, setup::Uid, EncryptionScheme, HashScheme, HashScheme, ChecksumScheme>, |
| 64 | LastVersionWins, |
| 65 | >; |
| 66 | |
| 67 | const ENV_ROOT: &str = "DIST_KILL9_ROOT"; |
| 68 | const ENV_START: &str = "DIST_KILL9_START"; |
| 69 | const ENV_FRESH: &str = "DIST_KILL9_FRESH"; |
| 70 | const ACK: &str = "ACK "; |
| 71 | |
| 72 | fn rid(n: u32) -> RecordId { |
| 73 | let mut bytes = [0xABu8; 32]; |
| 74 | bytes[..4].copy_from_slice(&n.to_be_bytes()); |
| 75 | RecordId::from_bytes(bytes) |
| 76 | } |
| 77 | |
| 78 | fn number(id: &RecordId) -> u32 { |
| 79 | let b = id.as_bytes(); |
| 80 | u32::from_be_bytes([b[0], b[1], b[2], b[3]]) |
| 81 | } |
| 82 | |
| 83 | // 200 to 1,500 bytes, some past 256 so that a width is exercised, and different for each n. |
| 84 | fn value(n: u32) -> Vec<u8> { |
| 85 | let len = 200 + (n % 97) as usize * 13; |
| 86 | (0..len).map(|i| (i as u32).wrapping_mul(n).wrapping_add(n >> 3) as u8).collect() |
| 87 | } |
| 88 | |
| 89 | fn schemes() -> Outcome<RestSchemesInput<EncryptionScheme, HashScheme, HashScheme, ChecksumScheme>> { |
| 90 | let enckey = [0x55u8; 32]; |
| 91 | let aes_gcm = res!(EncryptionScheme::new_aes_256_gcm_with_key(&enckey[..])); |
| 92 | Ok(RestSchemesInput::new( |
| 93 | Some(aes_gcm), |
| 94 | None::<HashScheme>, |
| 95 | None::<HashScheme>, |
| 96 | Some(ChecksumScheme::new_crc32()), |
| 97 | )) |
| 98 | } |
| 99 | |
| 100 | // One zone, a barrier on every write, and a cache far smaller than what is stored. The saved |
| 101 | // configuration wins on a reopen, so only a fresh root reads it. |
| 102 | fn one_zone() -> Outcome<OzoneConfig> { |
| 103 | let overrides = res!(mapdat!{ |
| 104 | 1u16 => mapdat!{ "dir" => "", "max_size" => 104_857_600u64 }, |
| 105 | }.get_map().ok_or_else(|| err!("The zone override map is not a map."; Test, Invalid))); |
| 106 | Ok(OzoneConfig { |
| 107 | num_zones: 1, |
| 108 | sync_on_write: true, |
| 109 | cache_size_limit_bytes: 4_096, |
| 110 | zone_overrides: overrides, |
| 111 | ..OzoneConfig::default() |
| 112 | }) |
| 113 | } |
| 114 | |
| 115 | fn open(root: &PathBuf, fresh: bool) -> Outcome<Eng> { |
| 116 | let oam = res!(OamConfig::new(1, 1)); |
| 117 | let dist = res!(DistOzoneConfig::new( |
| 118 | NodeId::from_bytes([1u8; 32]), |
| 119 | Vec::new(), |
| 120 | oam, |
| 121 | vec![res!(TableConfig::eventual("identity"))], |
| 122 | )); |
| 123 | let cfg = if fresh { Some(res!(one_zone())) } else { None }; |
| 124 | Eng::over_o3db(root.clone(), cfg, dist, res!(schemes()), setup::Uid::default(), LastVersionWins) |
| 125 | } |
| 126 | |
| 127 | fn root_dir(name: &str) -> PathBuf { |
| 128 | PathBuf::from(fmt!("./test_db_dist_kill9_{}", name)) |
| 129 | } |
| 130 | |
| 131 | // The child. It does nothing unless the parent set its environment, so an ordinary run of this |
| 132 | // binary passes it at once. |
| 133 | #[test] |
| 134 | fn kill9_child() -> Outcome<()> { |
| 135 | let root = match std::env::var(ENV_ROOT) { |
| 136 | Ok(r) => PathBuf::from(r), |
| 137 | Err(_) => return Ok(()), |
| 138 | }; |
| 139 | let start: u32 = res!(res!(std::env::var(ENV_START)).parse::<u32>()); |
| 140 | let fresh = std::env::var(ENV_FRESH).is_ok(); |
| 141 | // A child the parent never kills ends itself. |
| 142 | std::thread::spawn(|| { |
| 143 | std::thread::sleep(Duration::from_secs(150)); |
| 144 | std::process::exit(3); |
| 145 | }); |
| 146 | let e = res!(open(&root, fresh)); |
| 147 | for n in start.. { |
| 148 | let out = res!(e.put_at(Record::new(rid(n), "identity", value(n)), 1_000)); |
| 149 | assert!(out.local_persisted); |
| 150 | // Only now has the put returned, so only now is the id acknowledged. |
| 151 | println!("{}{}", ACK, n); |
| 152 | } |
| 153 | Ok(()) |
| 154 | } |
| 155 | |
| 156 | #[test] |
| 157 | fn kill9_acknowledged_puts_survive() -> Outcome<()> { |
| 158 | let root = root_dir("rounds"); |
| 159 | if root.exists() { |
| 160 | res!(std::fs::remove_dir_all(&root)); |
| 161 | } |
| 162 | res!(std::fs::create_dir_all(&root)); |
| 163 | let root = res!(root.canonicalize()); |
| 164 | let exe = res!(std::env::current_exe()); |
| 165 | |
| 166 | let mut acked: BTreeMap<u32, Vec<u8>> = BTreeMap::new(); |
| 167 | let rounds: [(u32, usize); 3] = [(1, 25), (1_000_000, 50), (2_000_000, 90)]; |
| 168 | for (round, (start, kill_after)) in rounds.iter().enumerate() { |
| 169 | let mut child = res!(Command::new(&exe) |
| 170 | .args(["kill9_child", "--exact", "--nocapture", "--test-threads=1"]) |
| 171 | .env(ENV_ROOT, &root) |
| 172 | .env(ENV_START, start.to_string()) |
| 173 | .envs(if round == 0 { Some((ENV_FRESH, "1")) } else { None }) |
| 174 | .stdout(Stdio::piped()) |
| 175 | .stderr(Stdio::inherit()) |
| 176 | .spawn()); |
| 177 | let out = res!(child.stdout.take().ok_or_else(|| err!("The child has no stdout."; Test, Missing))); |
| 178 | let mut lines = BufReader::new(out); |
| 179 | let mut this_round = 0usize; |
| 180 | let mut line = String::new(); |
| 181 | let mut killed = false; |
| 182 | loop { |
| 183 | line.clear(); |
| 184 | let got = res!(lines.read_line(&mut line)); |
| 185 | if got == 0 { |
| 186 | break; |
| 187 | } |
| 188 | // A line cut by the kill has no newline and is not an acknowledgement. |
| 189 | if !line.ends_with('\n') { |
| 190 | continue; |
| 191 | } |
| 192 | if let Some(n) = line.trim_end().strip_prefix(ACK) { |
| 193 | let n: u32 = res!(n.parse::<u32>()); |
| 194 | acked.insert(n, value(n)); |
| 195 | this_round += 1; |
| 196 | if this_round == *kill_after && !killed { |
| 197 | // SIGKILL, in mid-stream: the child is already inside its next put. |
| 198 | res!(child.kill()); |
| 199 | killed = true; |
| 200 | } |
| 201 | } |
| 202 | } |
| 203 | let status = res!(child.wait()); |
| 204 | assert!(killed, "round {}: the child ended after {} acknowledgements, before the kill ({:?})", |
| 205 | round, this_round, status); |
| 206 | assert_eq!(status.signal(), Some(9), "round {}: the child was not ended by SIGKILL", round); |
| 207 | assert!(this_round >= *kill_after); |
| 208 | |
| 209 | // Reopen what the crash left, and read everything acknowledged so far. |
| 210 | let e = res!(open(&root, false)); |
| 211 | for (n, want) in &acked { |
| 212 | let got = res!(e.storage().get("identity", &rid(*n))); |
| 213 | let got = res!(got.ok_or_else(|| err!("round {}: acknowledged record {} is missing.", round, n; Test, Missing))); |
| 214 | assert_eq!(&got.value, want, "round {}: record {} came back at other bytes", round, n); |
| 215 | } |
| 216 | // A write the child was inside when it died may be present, whole, or absent. |
| 217 | let digests = res!(e.storage().digests("identity")); |
| 218 | for d in &digests { |
| 219 | let n = number(&d.id); |
| 220 | if acked.contains_key(&n) { |
| 221 | continue; |
| 222 | } |
| 223 | if let Some(r) = res!(e.storage().get("identity", &d.id)) { |
| 224 | assert_eq!(r.value, value(n), "round {}: unacknowledged record {} is torn", round, n); |
| 225 | } |
| 226 | } |
| 227 | assert!(digests.len() >= acked.len()); |
| 228 | assert!(digests.len() <= acked.len() + round + 1, |
| 229 | "round {}: {} records stored, {} acknowledged", round, digests.len(), acked.len()); |
| 230 | res!(e.close()); |
| 231 | } |
| 232 | Ok(()) |
| 233 | } |