fix: canonical port key, remove()/restore() drift, escape-aware templates

Port keys no longer drift on zero-padding (host:080 vs host:80 collapsed
one canonical key), so a get() -> burn() pairing on a padded port no
longer raises a false "not in pool". Constructor/from_file now dedupe by
canonical key like add(). remove() decrements the rotation cursor when
the removed slot precedes it, fixing a skip/double-serve. restore() on
an unknown key now warns instead of silently no-op, matching remove()'s
contract. A genuine timed burn() under cooldown>0 is now tracked
separately from routine cooldown resting, so it logs WARNING instead of
being buried as DEBUG. Source validation uses is-not-None instead of
truthiness, so a falsy-but-provided source (e.g. template="") is
accepted. Bare "{}" template normalization is now escape-aware, so an
intentional "{{}}" literal survives instead of being corrupted.

Stored-shape change: burn()/cooldown deadlines now write int(time.time())
instead of float; comparisons are unaffected.

Docstrings compressed (module + next()); no behavior change.

Signed-off-by: disqualifier <dev@disqualifier.me>
This commit is contained in:
2026-07-02 23:25:52 -04:00
parent eef4b25f07
commit 7986a25fdc
5 changed files with 156 additions and 75 deletions
+47 -9
View File
@@ -9,15 +9,15 @@ edits. **Credentials are always injected — never hardcoded.**
## Install
```
aioproxies @ git+ssh://git@git.rethinkstudios.io/rethink-public/aioproxies.git@v0.2.2
aioproxies @ git+ssh://git@git.rethinkstudios.io/rethink-public/aioproxies.git@v0.3.0
# network helpers (current_ip / reset) need the extra:
aioproxies[net] @ git+ssh://git@git.rethinkstudios.io/rethink-public/aioproxies.git@v0.2.2
aioproxies[net] @ git+ssh://git@git.rethinkstudios.io/rethink-public/aioproxies.git@v0.3.0
```
The core has no dependencies. The `net` extra adds `aiohttp` for `current_ip` /
`reset`.
Drop the `@v0.2.2` suffix from the line above to install the latest unpinned.
Drop the `@v0.3.0` suffix from the line above to install the latest unpinned.
## Formatting
@@ -144,9 +144,11 @@ pm.is_burned(proxy) # current state (expired timed burns read False)
`burn`/`restore`/`is_burned`/`remove` accept **any proxy shape** — a spec string, a
`Proxy`, an aiohttp/camoufox/socks5 dict, or a url — all resolve to the same canonical
key (`host:port:user:pass`, or `host:port` auth-less). The password is part of the key,
so two proxies differing only by password are distinct slots. `burn` on a proxy not in
the pool raises `ValueError`.
key (`host:port:user:pass`, or `host:port` auth-less; the port is normalized so
`host:080` and `host:80` are one slot). The password is part of the key, so two proxies
differing only by password are distinct slots. `burn` on a proxy not in the pool raises
`ValueError`; `restore` on a proxy not in the pool logs a warning and no-ops (matching
`remove`'s contract, as of v0.3.0 — previously it silently did nothing with no signal).
### Cooldown
@@ -178,9 +180,12 @@ pm.remove(proxy) # drop a slot entirely (any shape) —
`replace` resets the rotation index and honors the manager's `shuffle` setting on the
incoming list. `remove` differs from `burn`: burn = unusable but still tracked; remove =
gone from the pool. Like the burn family, `add`/`replace` accept **any proxy shape**
(spec/`Proxy`/url/aiohttp dict/camoufox/socks5 dict). `canonical_key(shape)` and
`to_proxy(shape)` are exported if you need the key or a normalized `Proxy` yourself.
gone from the pool — removing a slot that precedes the rotation cursor adjusts the
cursor so `next()` doesn't skip a proxy. Like the burn family, `add`/`replace` accept
**any proxy shape** (spec/`Proxy`/url/aiohttp dict/camoufox/socks5 dict). The
constructor and `from_file` also dedupe by canonical key, same as `add`.
`canonical_key(shape)` and `to_proxy(shape)` are exported if you need the key or a
normalized `Proxy` yourself.
## Network helpers (optional)
@@ -204,6 +209,39 @@ await reset("https://provider/reset-url") # rotate upstream ip
## Changelog
### v0.3.0
- **Stored-shape change: int unix deadlines.** `burn(proxy, seconds)` and the
cooldown timeout write `int(time.time()) + n` instead of a `float`. Comparisons
(`is_burned`/`stats`/selection) are unaffected — this only tightens what gets
persisted into per-proxy state.
- **Port key normalization:** `key()` strips leading zeros from the port, so
`host:080` and `host:80` are the same canonical slot. Previously a zero-padded
port could break a `get()``burn()` pairing (the burn would raise "not in
pool" against the proxy that was just handed out).
- **Constructor / `from_file` dedupe by canonical key**, matching `add()`. A
proxy list with repeated entries (e.g. same host:port:user:pass twice) no
longer inflates the pool or double-weights rotation.
- **`remove()` no longer skews rotation.** Removing a proxy that precedes the
rotation cursor now decrements the cursor, so the next `next()` call doesn't
skip or double-serve a proxy.
- **`restore()` on an unknown proxy now logs a warning and no-ops**, matching
`remove()`'s contract (previously it silently did nothing, with `burn()`
raising for the same precondition and `remove()` warning — `restore()` was the
odd one out).
- **Genuine timed burns under `cooldown>0` now log at WARNING**, not DEBUG. A
real `burn(proxy, seconds)` is now tracked separately from the manager's own
cooldown resting, so it no longer gets buried as routine cooldown noise when
every proxy in the pool happens to also be cooling down.
- **Source truthiness → presence.** The constructor's exactly-one-source check
now uses `is not None` instead of truthiness, so `template=""` (or another
falsy-but-explicitly-provided source) is accepted rather than silently
rejected as "no source given."
- **Escape-aware bare-`{}` template normalization.** A template's bare `{}` is
still filled as the session slot, but an escaped `{{}}` (str.format's own
convention for a literal `{}` in the output) now survives untouched instead of
being corrupted into `{{session}}`.
### v0.2.1
- **Legible missing-template-field error:** a `template=` placeholder not supplied to
+1 -1
View File
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
[project]
name = "aioproxies"
version = "0.2.2"
version = "0.3.0"
description = "proxy parsing, formatting, health, and pool management for aiohttp/aioweb, camoufox, and socks5"
requires-python = ">=3.10"
dependencies = []
+1 -1
View File
@@ -18,4 +18,4 @@ __all__ = [
"to_proxy",
]
__version__ = "0.2.2"
__version__ = "0.3.0"
+91 -59
View File
@@ -1,35 +1,35 @@
"""proxy source management: session templates, rotation, static.
`AioProxies` is constructed with exactly one source and hands out `Proxy`
objects. it carries no module-level globals (rotation state is per-instance) and
never exits the process — a missing proxy file raises, it does not `sys.exit`.
`AioProxies` is constructed with exactly one source and hands out `Proxy` objects.
no module-level globals (rotation state is per-instance); a missing proxy file
raises, never `sys.exit`.
sources:
- template: a format string with named placeholders. `{session}` is filled with
a fresh session id on each `next()`; any other placeholder (e.g. `{country}`,
`{ttl}`) is filled from keyword args passed to `next(**fields)`. a bare `{}` is
also accepted and treated as the session slot (back-compat with simple templates).
- proxies: a list of specs cycled round-robin
- template: a format string. `{session}` is filled with a fresh id on each
`next()`; other placeholders (`{country}`, `{ttl}`, ...) come from
`next(**fields)`. a bare `{}` is treated as the session slot unless escaped as
`{{}}` (back-compat with simple templates).
- proxies: a list of specs cycled round-robin (deduped by canonical key)
- static: one fixed proxy
v0.2.0 adds proxy health to the rotating list source only — burn/timeout, usage
counters, reuse cooldown, and pool management (replace/add/remove). these are
keyed by each proxy's canonical key (`host:port:user:pass`, or `host:port` for
auth-less / IP-authenticated proxies). on template/static sources they are no-ops
that log a warning and return cleanly, so generic caller code can call them
regardless of source.
proxy health (rotating list source only) — burn/timeout, usage counters, reuse
cooldown, pool edits (replace/add/remove) — keyed by each proxy's canonical key
(`host:port:user:pass`, or `host:port` auth-less; port is normalized so `:080`
and `:80` collapse to one key). on template/static sources these are no-ops that
log a warning and return, so generic caller code can call them regardless of
source.
per-proxy state (keyed by canonical key):
- `uses`: pure counter, incremented on every handout. never drives selection;
survives burns (a proxy can read "used 500x and dead").
- `timeout`: availability state. `None`/`0` = fine; `-1` = dead/permanent (manual
restore only); a future unix ts = timed out until then (lazy, checked against
`time.time()`, no timers). cooldown and timed burns share this field. durations
(seconds) only ever exist as arguments — converted to `now + seconds` and
discarded; the lib never stores a raw duration.
per-proxy state (keyed by canonical key): `uses` is a pure counter, incremented on
every handout, never drives selection, survives burns. `timeout` is the
availability state — `None`/`0` fine, `-1` dead/permanent (manual restore only), a
future unix ts = timed out until then. this core is sync: no timers, lazy expiry
checked only against `time.time()` at call time. cooldown and timed burns share
this field; durations only ever exist as call arguments, converted to `now +
seconds` and discarded — a raw duration is never stored.
"""
import logging
import random
import re
import string
import time
from typing import Dict, List, Optional, Union
@@ -39,6 +39,7 @@ from .proxy import Proxy, canonical_key, parse, to_proxy
log = logging.getLogger(__name__)
_DEAD = -1
_BARE_SESSION_SLOT = re.compile(r"(?<!\{)\{\}(?!\})")
class ProxiesExhaustedError(Exception):
@@ -60,22 +61,46 @@ class AioProxies:
):
if proxies is not None and len(proxies) == 0:
raise ValueError("proxies list is empty; provide at least one proxy")
sources = [s for s in (template, proxies, static) if s]
sources = [s for s in (template, proxies, static) if s is not None]
if len(sources) != 1:
raise ValueError("provide exactly one of: template, proxies, static")
self.template = template.replace("{}", "{session}") if template else None
self.template = self._normalize_template(template) if template is not None else None
self.session_len = session_len
self.cooldown = cooldown
self._shuffle = shuffle
self._static = parse(static) if static else None
self._proxies = [parse(p) for p in proxies] if proxies else []
self._static = parse(static) if static is not None else None
self._proxies = self._dedupe([parse(p) for p in proxies]) if proxies is not None else []
if self._proxies and shuffle:
random.shuffle(self._proxies)
self._state: Dict[str, Dict[str, object]] = {}
for proxy in self._proxies:
self._state.setdefault(proxy.key(), {"uses": 0, "timeout": None})
self._state.setdefault(proxy.key(), self._fresh_state())
self._index = 0
@staticmethod
def _dedupe(proxies: List[Proxy]) -> List[Proxy]:
"""drop later entries sharing a canonical key, preserving first-seen order"""
seen: Dict[str, Proxy] = {}
for proxy in proxies:
seen.setdefault(proxy.key(), proxy)
return list(seen.values())
@staticmethod
def _fresh_state() -> Dict[str, object]:
"""a clean per-proxy state entry: no uses, no timeout, not burned"""
return {"uses": 0, "timeout": None, "burned": False}
@staticmethod
def _normalize_template(template: str) -> str:
"""fill a bare `{}` session slot, escape-aware
a lone `{}` is treated as `{session}` (back-compat with simple templates).
an escaped literal `{{}}` — str.format's own convention for a literal `{}`
in the output — is left untouched so it survives to render as `{}`, not
`{session}`.
"""
return _BARE_SESSION_SLOT.sub("{session}", template)
@classmethod
def from_file(cls, path: str, **kwargs) -> "AioProxies":
"""build a rotating manager from a newline-delimited proxy file
@@ -112,15 +137,14 @@ class AioProxies:
def next(self, **fields: object) -> Proxy:
"""return the next proxy from the configured source
for template sources, `{session}` is always filled with a fresh id and any
other named placeholder is filled from `fields` (e.g. next(country="ca",
ttl=30)). fields are ignored by list/static sources.
template sources: `{session}` is always filled with a fresh id, other
placeholders come from `fields` (e.g. next(country="ca", ttl=30)). fields
are ignored by list/static sources.
for list sources, skips proxies whose timeout is active (-1 dead, or a
future ts not yet passed), increments `uses` on handout, and applies the
manager's cooldown. if none are fine but some are merely timed, hands out
the one recovering soonest (with a warning); raises ProxiesExhaustedError
if every proxy is permanently dead.
list sources: skips proxies whose timeout is active (-1 dead, or a future
ts not yet passed), increments `uses` on handout, applies the manager's
cooldown. if none are fine but some are merely timed, hands out the one
recovering soonest (warns); raises ProxiesExhaustedError if all are dead.
"""
if self._static is not None:
return self._static
@@ -172,24 +196,19 @@ class AioProxies:
state = self._state[proxy.key()]
state["uses"] = int(state["uses"]) + 1
if self.cooldown > 0:
state["timeout"] = now + self.cooldown
state["timeout"] = int(now) + self.cooldown
return proxy
def _has_burned(self) -> bool:
"""whether the empty 'fine' tier reflects a genuine burn, not just cooldown
"""whether the empty 'fine' tier reflects a genuine burn(), not just cooldown
a dead (-1) proxy is always a real burn. a future-ts timeout is a real timed
burn only when cooldown is off; with cooldown on, future-ts entries are the
manager's own resting and not a pool-health signal. used to pick warning
(real trouble) vs debug (normal cooldown) on the soonest-recovering path.
tracked directly via each state's `burned` flag (set by burn(), cleared by
restore()) so a real burn is never masked by the manager's own cooldown
resting on the same `timeout` field, regardless of whether cooldown is on.
used to pick warning (real trouble) vs debug (normal cooldown) on the
soonest-recovering path.
"""
for state in self._state.values():
timeout = state["timeout"]
if timeout == _DEAD:
return True
if self.cooldown == 0 and timeout not in (None, 0):
return True
return False
return any(state["burned"] for state in self._state.values())
def _soonest_recovering(self) -> Optional[int]:
"""index of the timed (non-dead) proxy recovering soonest, or None"""
@@ -224,16 +243,24 @@ class AioProxies:
if seconds is None:
self._state[key]["timeout"] = _DEAD
else:
self._state[key]["timeout"] = time.time() + seconds
self._state[key]["timeout"] = int(time.time()) + seconds
self._state[key]["burned"] = True
def restore(self, proxy: Union[str, Proxy, Dict[str, str]]) -> None:
"""clear any burn/timeout on a proxy (back to fine). no-op if already fine"""
"""clear any burn/timeout on a proxy (back to fine). no-op if already fine
matches remove()'s contract for an unknown proxy: logs a warning and returns
rather than silently doing nothing.
"""
if not self._is_list_source():
self._warn_non_list("restore()")
return
key = canonical_key(proxy)
if key in self._state:
self._state[key]["timeout"] = None
if key not in self._state:
log.warning("restore(): proxy not in pool: %s", key)
return
self._state[key]["timeout"] = None
self._state[key]["burned"] = False
def is_burned(self, proxy: Union[str, Proxy, Dict[str, str]]) -> bool:
"""whether a proxy is currently unavailable (lazy expiry of timed burns)
@@ -310,7 +337,7 @@ class AioProxies:
if keep_state and key in old_state:
new_state[key] = old_state[key]
else:
new_state[key] = {"uses": 0, "timeout": None}
new_state[key] = self._fresh_state()
self._proxies = new_proxies
self._state = new_state
self._index = 0
@@ -327,13 +354,15 @@ class AioProxies:
if key in self._state:
continue
self._proxies.append(proxy)
self._state[key] = {"uses": 0, "timeout": None}
self._state[key] = self._fresh_state()
def remove(self, proxy: Union[str, Proxy, Dict[str, str]]) -> None:
"""drop a proxy from the pool entirely (by canonical key, any shape)
distinct from burn (burn = unusable but tracked; remove = gone). clamps the
rotation index if needed. no-op + warning if the proxy is not present.
distinct from burn (burn = unusable but tracked; remove = gone). decrements
the rotation index when the removed slot precedes it, so the next() cursor
still lands on the same upcoming proxy instead of skipping one. no-op +
warning if the proxy is not present.
"""
if not self._is_list_source():
self._warn_non_list("remove()")
@@ -342,12 +371,15 @@ class AioProxies:
if key not in self._state:
log.warning("remove(): proxy not in pool: %s", key)
return
removed_idx = next(i for i, p in enumerate(self._proxies) if p.key() == key)
self._proxies = [p for p in self._proxies if p.key() != key]
del self._state[key]
if self._proxies:
self._index %= len(self._proxies)
else:
if not self._proxies:
self._index = 0
else:
if removed_idx < self._index:
self._index -= 1
self._index %= len(self._proxies)
# name aliases — same class, call it whichever reads best at your call site
+16 -5
View File
@@ -19,6 +19,15 @@ SCHEME_HTTP = "http"
SCHEME_SOCKS5 = "socks5"
def normalize_port(port: Union[str, int]) -> str:
"""canonical port string: parsed to int, rendered without zero-padding
keeps host:080 and host:80 as one canonical key. raises ValueError on a
non-integer port rather than silently keying it as-is.
"""
return str(int(port))
@dataclass
class Proxy:
"""a single proxy endpoint with optional auth"""
@@ -38,14 +47,16 @@ class Proxy:
the host is lowercased (hostnames are case-insensitive per DNS, so
PROXY.example.com and proxy.example.com are the same host and collapse to one
key); port and credentials are kept verbatim. the password is included in
full (two proxies differing only by password are distinct slots). auth-less
proxies collapse to host:port with no trailing colons.
key); credentials are kept verbatim. the port is normalized (leading zeros
stripped) so host:080 and host:80 collapse to one key. the password is
included in full (two proxies differing only by password are distinct
slots). auth-less proxies collapse to host:port with no trailing colons.
"""
host = self.host.lower()
port = normalize_port(self.port)
if self.has_auth:
return f"{host}:{self.port}:{self.user}:{self.password}"
return f"{host}:{self.port}"
return f"{host}:{port}:{self.user}:{self.password}"
return f"{host}:{port}"
def url(self, scheme: str = SCHEME_HTTP) -> str:
"""render as a url, embedding auth when present