Review follow-ups to 73a93f1.
Bound in-flight calls. rpcclient's single sendPostHandler goroutine
imposed an accidental ceiling of one concurrent RPC; removing it without
putting anything in its place left no bound at all. grpc-go supplies
none either -- this server sets no MaxConcurrentStreams, so the default
is math.MaxUint32 -- and dragonxd answers RPC with 8 worker threads
behind a 4096-deep queue, shared on the pool node with getblocktemplate.
Overload would therefore surface as mining latency rather than as an
error we could back off on. MaxConnsPerHost blocks the caller at the
limit instead of dialling more, which is the backpressure wanted;
MaxIdleConnsPerHost alone would only cap reuse and let us exceed the
limit while churning connections. Default 8, matching the node's
DEFAULT_HTTP_THREADS, tunable with -rpc-max-concurrent. Even 8 removes
all of the head-of-line blocking this work set out to fix.
IdleConnTimeout 90s -> 20s. dragonxd closes idle connections at 30s
(DEFAULT_HTTP_SERVER_TIMEOUT, applied via evhttp_set_timeout and not
overridden in DRAGONX.conf). At 90s we were always the second to close,
so a request could be written into a connection the server had already
sent a FIN for, and Go will not retry a POST once bytes are on the wire.
Closing first removes the race.
Reject a negative -rpc-timeout, which silently meant "unbounded", the
same as the documented 0. The check has to run after flag.Parse(); it
was initially placed before it and never fired.
Also correct the coinsupply note: hush_coinsupply walks the block index
back to genesis, loading each block from disk and memoising newcoins and
zfunds into the CBlockIndex, so the first call pays for the whole chain
and later ones are nearly free. It is not a UTXO-set scan, as the
earlier comment claimed. The measured 48s/3s figures are unchanged.
Verified: five concurrent GetLightdInfo calls all return grpc-status 0
with no errors, 46 blocks ingested, and the daemon holds 2 sockets to
the node rather than one per request; a negative timeout exits 1.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FU87LdsJZiZkfq1eXubpeo
lightwalletd used github.com/btcsuite/btcd/rpcclient for exactly one
method, RawRequest, and that package has two defects that together
produce the failure seen on the primary node:
1. It has no timeout and no way to set one. The http.Client is built
in an unexported newHTTPClient() and ConnConfig exposes no Timeout
field, so a call can hang forever. Calls were observed running past
five minutes against a healthy node that answered the same query
from dragonx-cli in 2ms.
2. In HTTP POST mode it runs ONE sendPostHandler goroutine which calls
handleSendPostMessage synchronously, so the whole process has at
most one RPC in flight. One stuck call therefore blocks the block
ingestor, the mempool monitor and every user-facing gRPC handler at
once.
These compound: a timeout alone would not have been enough, because
bounding the caller's wait still leaves the shared goroutine stuck
inside http.Client.Do with everything queued behind it. Only a timeout
on the HTTP client aborts the in-flight request, and only dropping the
shared goroutine lets independent callers proceed.
Patching the vendored copy is not an option here: go.mod declares
go 1.12, so automatic vendor mode (go >= 1.14) is off and the committed
vendor/ tree is silently ignored in favour of the module cache. It is
also stale -- vendor/modules.txt disagrees with go.mod on btcd,
protobuf, logrus, sqlite3 and six other modules, so -mod=vendor cannot
build at all, and `go mod vendor` would discard any patch.
Replace it with package zrpc: a ~150-line JSON-RPC client that is safe
for concurrent use and takes a timeout. The request envelope, ID
sequence, basic auth and error semantics are deliberately identical.
RPCError.Error() still renders as "<code>: <message>" because callers
recover the numeric code from the string -- common.GetSaplingInfo
checks for -8 via strings.SplitN(err.Error(), ":", 2) -- and the
non-JSON body path still reports `status code: %d, response: %q`.
Default timeout 120s, tunable with -rpc-timeout (0 disables). The bound
is set by the slowest legitimate call: coinsupply measured 48s against
a cold UTXO set, 3s once cached, so a tighter timeout would turn a
slow-but-working call into a hard failure.
Also drops rpcclient's Close=true, which opened a fresh TCP connection
per request and left hundreds of sockets in TIME_WAIT on a busy node;
idle connections are now reused and capped.
Verified against a live dragonxd: getblockchaininfo returns chain=main;
a bad height yields exactly "-8: Block height out of range" and parses
back to -8; a 1ns timeout aborts in 54us instead of hanging; and the
built daemon serves GetLightdInfo at the current height with no errors.
The pre-existing parser TestCompactBlocks failure is unrelated and
reproduces on the base commit.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FU87LdsJZiZkfq1eXubpeo
Bump for the GetTransaction crash fix, and make the version one value
instead of two literals that could drift.
It was duplicated: cmd/server/main.go had `var version = "0.1.1"` for
--version, and frontend/service.go had "0.1.1-dragonxlightd" inline in
the LightdInfo reply. The gRPC one is the load-bearing copy -- it is
walletrpc/service.proto:48, so every client reads it, and it is the only
way to tell from off-box which build a node is running.
That property is the point of bumping now rather than later. With it, a
rollout can be verified by probing each endpoint over TLS and reading
the advertised version, instead of shelling in to compare binary
checksums, and instead of the only alternative positive test -- calling
GetTransaction on a mempool txid, which proves the fix by crashing any
node that does not have it.
Verified: --version prints 0.1.2, and a GetLightdInfo probe against a
test instance returns 0.1.2-dragonxlightd where production still returns
0.1.1-dragonxlightd.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FU87LdsJZiZkfq1eXubpeo
Adaptive-lag: advertise a tip that trails the real tip by a reorg-rate-driven
lag (GetLatestBlock/GetLightdInfo) so wallets anchor shielded spends at a
settled height during reorg churn. Configurable via
-adaptive-lag/-lag-min/-lag-max/-lag-window; monitor_lwd.sh runs
-lag-min 4 -lag-max 12 -lag-window 30.
Startup coinsupply: the informational startup coinsupply RPC (result used only
for a log line) blocked the gRPC bind for minutes on a node whose supply index
is cold, keeping the lite endpoint down on every restart while it also starved
the block-cache ingestor. Run it in a goroutine so the bind is never blocked;
clients still get coinsupply on demand via the GetCoinsupply RPC.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
We should also be consistent in the SQLite database about either storing
as TEXT in display order, or storing as BLOB in wire order. But as
that's another breaking change, let's bundle it in with anything else we
need to do to address this problem.
