Struct DebugState 

pub struct DebugState {

    pub enabled: bool,
    pub sandbox_enabled: bool,
    pub trigger_header: String,
    pub trace_all: bool,
    pub capture_bodies: bool,
    pub max_body_bytes: usize,
    pub max_traces: usize,
    pub max_steps: usize,
    pub sandbox_timeout_seconds: u64,
    redaction: RedactionPolicy,
    traces: Mutex<VecDeque<Arc<Trace>>>,
    seq: AtomicU64,
    evicted: AtomicU64,
}

Resolved debug settings plus the trace buffer.

Every field is immutable after construction — system.yaml is not hot-reloaded, so debug mode cannot be switched on in a running process. That is deliberate: it means a compromised Admin API credential cannot start capturing request contexts.

Fields

enabled: bool

sandbox_enabled: bool

trigger_header: String

Lowercased, because GatewayRequest stores header names as hyper normalised them.

trace_all: bool

capture_bodies: bool

max_body_bytes: usize

max_traces: usize

max_steps: usize

sandbox_timeout_seconds: u64

redaction: RedactionPolicy

traces: Mutex<VecDeque<Arc<Trace>>>

seq: AtomicU64

evicted: AtomicU64

Traces dropped to keep the buffer at max_traces.

Kept so a caller can tell an empty result apart from a rotated-out one. Without it, filtering for something that has just aged out is indistinguishable from filtering for something that never happened – which has already produced one confident, wrong “the filter is broken” diagnosis.

Implementations

impl DebugState

pub fn new(cfg: &DebugConfig) -> Self

Builds the runtime state from config.

pub fn capture_options(&self) -> CaptureOptions

Capture knobs for a snapshot.

pub fn should_trace(&self, headers: &HashMap<String, Vec<String>>) -> bool

Whether this request should be traced.

The whole cost on an untraced request: one HashMap::get against the already-built header map, and only when debug is enabled at all.

pub fn header_opt_in(&self, headers: &HashMap<String, Vec<String>>) -> bool

Whether the request explicitly asked to be traced via the trigger header, as opposed to being swept up by trace_all.

The x-featherbit-trace-id response header is returned only in this case: the id is a reply to a caller who opted in. Under trace_all the traffic is anonymous, so stamping every response with a debug id would leak it to clients that never asked — those traces are found in the panel/list instead.

pub fn next_seq(&self) -> u64

Next monotonic sequence number.

pub fn record(&self, trace: Trace)

Stores a trace, evicting the oldest entries beyond max_traces.

pub fn retention(&self) -> Retention

What the buffer holds and what it has discarded.

Returned alongside every trace listing so an empty result is self-explanatory: with truncated set and an oldest_seq to compare against, “no traces matched” and “the matches aged out” stop looking the same.

pub fn list(&self) -> Vec<TraceSummary>

Summaries, newest first.

pub fn get(&self, id: &str) -> Option<Arc<Trace>>

Fetches one trace. Clones an Arc under the lock so serialization happens outside it.

pub fn clear(&self) -> usize

Empties the buffer, returning how many traces were dropped.

pub fn len(&self) -> usize

Number of traces currently held.