igneum-network/igneum
1
0
Fork 0
Code Issues Pull requests Releases Activity

Pool daemon: --template-parallel (default 2) bounds the template fetches in flight against the node (the ten-member window on pool-1, 7 October 2026: ten parallel getBlockTemplate calls on one gRPC connection queued past the client's request timeout from 12:54Z, no job for 36 minutes); pool.md 10.5a records the window's two stacked faults

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
igneum-labs 2026-10-07 13:38:52 +00:00
parent 39d48dfe64
commit ada78f8c45
6 changed files with 30 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
docs/plans/pool.md
View file

@ -461,6 +461,20 @@ Consequences by tier, from these numbers (the standing rule of 5 October):
| A pool user without a node | cannot join the open pool; pool-0 as before | the page says so |
| The network | the open pool's stale rate was 9.6 percent at 2 shares a second with ten daemons on one host; a public chain at the same rate across the internet loses more to forks, which is the uncles row (10.6); every block's coinbase grew by 48 bytes per payee, at most 256 payees, under the devnet's 16,384-byte cap | the uncles row stays open; the public chain's rate is O-9.10 |
### 10.5a The ten-member window on pool-1 (the fleet lane, 7 October 2026, 12:37Z to 13:30Z)
Ten rented RTX 3070 members (`mine none`, no node of their own) on the fixed pair, daemon 03457d96 and miner 9829bdf7,
pool-1's node 5899f603, plain TCP. Result: 10 of 10 authorized, 0 shares, 0 blocks, 0 payouts, every member's vardiff
easing from shift 10 to 27 with nothing to measure (bundle `~/igneum-fleet/pool-run-1328/` on the fleet lane's side). Two
faults stacked, each its own row:
| Fault | Side | What the logs show | Fix |
|---|---|---|---|
| The daemon got no template after 12:54Z | daemon (this lane) | 1,891 "template for member N: RPC request timeout" lines, STATUS "NODE NOT ANSWERING: last template 2,072 s ago" at 13:28Z, while the node read synced with 3 to 4 peers and answered its own miner's 578 template requests | ten parallel `getBlockTemplate` calls on one gRPC connection against a node that builds a template in 1.6 to 1.8 s queued past the client's own request timeout; `--template-parallel` (default 2) bounds the fetches in flight, the rest take the next tick |
| The members' NVRTC prepare never completed across the epoch boundary 78 to 79 | the member's worker path | every member printed `prepare started ... (NVRTC sm_86 in the background)` and never `prepared` or `prepare-failed`; the worker sat at 0 percent on the old pack; a solo miner on a standing box crossed the same boundary through its exit-42-and-re-export path, which the member path lacks | the register's row (docs/plans/miner-faults.md, the reliability lane's number): a member that hears nothing for 120 s after a prepare treats it as failed, re-exports, re-sends once, then restarts its worker with the reason shown; the pool flags a member with 0 accepted across an epoch roll |
So the window holds no pool-0 share numbers; its rerun waits for both fixes on the 0.3.21 pair.
### 10.6 Open after this round
| Item | Why it is open | What closes it |

1
pool/README.md
View file

@ -54,6 +54,7 @@ igneum-pool --node grpc://127.0.0.1:26610 --evm-rpc http://127.0.0.1:26790 \
| `--stale-grace-ms` | `2000` | a share on a superseded job inside the grace still pays |
| `--orphan-after-daa` | `120` | a pending block this far behind the virtual and not blue is an orphan |
| `--verify-threads` | `2` | share checks running at once on the CPU |
| `--template-parallel` | `2` | template fetches in flight against the node (one gRPC connection is one request stream; a burst of one fetch per member against a node that builds a template in over a second queued past the client's request timeout on pool-1, 7 October 2026) |
| `--tls-cert`, `--tls-key` | none | TLS 1.3 on the member port (spec 9.3) from a PEM chain and key |
| `--tls-self-signed` | off | TLS 1.3 from a self-signed pair made under the data dir on first start; the pool prints the certificate's pin, members pass it as `--pool-pin` |
| `--allow-plain` | off | plain TCP members on testnet or mainnet (the devnet allows plain by default; the public networks refuse to start in the clear without this) |

9
pool/src/config.rs
View file

@ -39,6 +39,11 @@ pub struct Config {
/// its own miner plus four members, the fixed 5 s gave up on the parallel fetches, no job was issued for 7 minutes,
/// every block the stale templates found was an orphan, and vardiff could not act (a new target rides with a job)
pub template_timeout_s: u64,
/// Template fetches in flight at once against the node (7 October 2026, the ten-member window on pool-1: ten
/// parallel getBlockTemplate calls on one gRPC connection against a node that builds a template in 1.6 to 1.8 s
/// queued past the client's own request timeout, "RPC request timeout" on every member template from 12:54Z for
/// 36 minutes, no job issued, no share); the rest of the members take the next tick's template
pub template_parallel: usize,
/// TLS 1.3 on the member port (spec 9.3, `tls.rs`): a PEM chain and key, or a self-signed pair under the data dir
pub tls_cert: Option<PathBuf>,
pub tls_key: Option<PathBuf>,
@ -68,7 +73,7 @@ pub struct Config {
impl Config {
pub fn usage() -> ! {
eprintln!(
"usage: igneum-pool [--node grpc://127.0.0.1:26610]... [--evm-rpc http://127.0.0.1:26790] [--listen 0.0.0.0:4463] [--node-wait-secs 60] [--template-timeout-s 20]\n\
"usage: igneum-pool [--node grpc://127.0.0.1:26610]... [--evm-rpc http://127.0.0.1:26790] [--listen 0.0.0.0:4463] [--node-wait-secs 60] [--template-timeout-s 20] [--template-parallel 2]\n\
\x20 [--http 127.0.0.1:4480] [--data-dir ./data] [--payout-key <file>] [--fee-percent 1] [--min-payout 1.0]\n\
\x20 [--pplns-window 2.0] [--payout-interval-s 60] [--dry-run] [--network devnet|testnet|mainnet]\n\
\x20 [--share-interval-s 10] [--min-shift 0] [--max-shift 60] [--stale-grace-ms 2000] [--orphan-after-daa 120]\n\
@ -105,6 +110,7 @@ impl Config {
snapshot_interval_s: 15,
node_wait_secs: 60,
template_timeout_s: 20,
template_parallel: 2,
tls_cert: None,
tls_key: None,
tls_self_signed: false,
@ -151,6 +157,7 @@ impl Config {
"--snapshot-interval-s" => c.snapshot_interval_s = val().parse().unwrap_or_else(|_| Config::usage()),
"--node-wait-secs" => c.node_wait_secs = val().parse().unwrap_or_else(|_| Config::usage()),
"--template-timeout-s" => c.template_timeout_s = val().parse::<u64>().unwrap_or_else(|_| Config::usage()).max(1),
"--template-parallel" => c.template_parallel = val().parse::<usize>().unwrap_or_else(|_| Config::usage()).max(1),
"--tls-cert" => c.tls_cert = Some(PathBuf::from(val())),
"--tls-key" => c.tls_key = Some(PathBuf::from(val())),
"--tls-self-signed" => c.tls_self_signed = true,

1
pool/src/main.rs
View file

@ -179,6 +179,7 @@ async fn main() {
net: Mutex::new(NetInfo::default()),
want_templates: tokio::sync::Notify::new(),
verify_permits: tokio::sync::Semaphore::new(cfg.verify_threads),
template_permits: tokio::sync::Semaphore::new(cfg.template_parallel),
started: Instant::now(),
cfg: cfg.clone(),
template_failures: AtomicU64::new(0),

4
pool/src/node.rs
View file

@ -68,6 +68,10 @@ pub async fn fetch_job(pool: &Arc<Pool>, member: &Arc<Member>) -> Result<(), Str
Some(o) => o.base_extra_data(member),
None => extra_data(member, &pool.pool_address),
};
// at most `--template-parallel` fetches in flight: the gRPC client is one request stream per connection and a
// node under load builds a template in over a second, so a burst of one fetch per member queued past the client's
// request timeout (pool-1, 7 October 2026, 12:54Z to 13:30Z: 1,891 "RPC request timeout" lines, no job issued)
let _permit = pool.template_permits.acquire().await.map_err(|e| e.to_string())?;
let tmpl = match tokio::time::timeout(Duration::from_secs(pool.cfg.template_timeout_s), pool.node.get_block_template(pool.pay_address.clone(), base_extra)).await {
Ok(Ok(t)) => t,
Ok(Err(e)) => {

2
pool/src/pool.rs
View file

@ -123,6 +123,8 @@ pub struct Pool {
pub net: Mutex<NetInfo>,
pub want_templates: tokio::sync::Notify,
pub verify_permits: tokio::sync::Semaphore,
/// Template fetches in flight against the node (`--template-parallel`)
pub template_permits: tokio::sync::Semaphore,
pub started: Instant,
/// Template fetches that failed or timed out since start, and the unix ms of the last one that succeeded (0 =
/// never): a daemon whose node answers no template issues no job, and the STATUS line in the log says so