diff --git a/backend/app/services/collector.py b/backend/app/services/collector.py
index 6befd5c..77f47bc 100644
--- a/backend/app/services/collector.py
+++ b/backend/app/services/collector.py
@@ -1,6 +1,7 @@
 import asyncio
 import logging
 from datetime import datetime, timezone
+from random import uniform
 from sqlalchemy import select
 from sqlalchemy.ext.asyncio import AsyncSession
 from sqlalchemy.exc import SQLAlchemyError
@@ -15,6 +16,9 @@ logger = logging.getLogger(__name__)
 settings = get_settings()
 _failure_state: dict[int, dict[str, object]] = {}
 _failure_log_interval_seconds = 300
+_backoff_base_seconds = max(3, int(settings.poll_interval_seconds))
+_backoff_max_seconds = 300
+_backoff_jitter_factor = 0.15
 
 
 def build_target_dsn(target: Target) -> str:
@@ -181,38 +185,66 @@ async def collect_once() -> None:
     async with SessionLocal() as db:
         targets = (await db.scalars(select(Target))).all()
 
+    active_target_ids = {target.id for target in targets}
+    stale_target_ids = [target_id for target_id in _failure_state.keys() if target_id not in active_target_ids]
+    for stale_target_id in stale_target_ids:
+        _failure_state.pop(stale_target_id, None)
+
     for target in targets:
+        now = datetime.now(timezone.utc)
+        state = _failure_state.get(target.id)
+        if state:
+            next_attempt_at = state.get("next_attempt_at")
+            if isinstance(next_attempt_at, datetime) and now < next_attempt_at:
+                continue
+
         try:
             await collect_target(target)
             prev = _failure_state.pop(target.id, None)
             if prev:
+                first_failure_at = prev.get("first_failure_at")
+                downtime_seconds = None
+                if isinstance(first_failure_at, datetime):
+                    downtime_seconds = max(0, int((now - first_failure_at).total_seconds()))
                 logger.info(
-                    "collector_target_recovered target=%s after_failures=%s last_error=%s",
+                    "collector_target_recovered target=%s after_failures=%s downtime_seconds=%s last_error=%s",
                     target.id,
                     prev.get("count", 0),
+                    downtime_seconds,
                     prev.get("error"),
                 )
         except (OSError, SQLAlchemyError, asyncpg.PostgresError, Exception) as exc:
-            now = datetime.now(timezone.utc)
             current_error = str(exc)
             error_class = exc.__class__.__name__
             state = _failure_state.get(target.id)
             if state is None:
+                next_delay = min(_backoff_max_seconds, _backoff_base_seconds)
+                jitter = next_delay * _backoff_jitter_factor
+                next_delay = max(1, int(next_delay + uniform(-jitter, jitter)))
+                next_attempt_at = now.timestamp() + next_delay
                 _failure_state[target.id] = {
                     "count": 1,
+                    "first_failure_at": now,
                     "last_log_at": now,
                     "error": current_error,
+                    "next_attempt_at": datetime.fromtimestamp(next_attempt_at, tz=timezone.utc),
                 }
                 logger.warning(
-                    "collector_target_unreachable target=%s error_class=%s err=%s consecutive_failures=%s",
+                    "collector_target_unreachable target=%s error_class=%s err=%s consecutive_failures=%s retry_in_seconds=%s",
                     target.id,
                     error_class,
                     current_error,
                     1,
+                    next_delay,
                 )
                 continue
 
             count = int(state.get("count", 0)) + 1
+            raw_backoff = min(_backoff_max_seconds, _backoff_base_seconds * (2 ** min(count - 1, 10)))
+            jitter = raw_backoff * _backoff_jitter_factor
+            next_delay = max(1, int(raw_backoff + uniform(-jitter, jitter)))
+            state["next_attempt_at"] = datetime.fromtimestamp(now.timestamp() + next_delay, tz=timezone.utc)
+
             last_log_at = state.get("last_log_at")
             last_logged_error = str(state.get("error", ""))
             should_log = False
@@ -228,11 +260,12 @@ async def collect_once() -> None:
                 state["last_log_at"] = now
                 state["error"] = current_error
                 logger.warning(
-                    "collector_target_unreachable target=%s error_class=%s err=%s consecutive_failures=%s",
+                    "collector_target_unreachable target=%s error_class=%s err=%s consecutive_failures=%s retry_in_seconds=%s",
                     target.id,
                     error_class,
                     current_error,
                     count,
+                    next_delay,
                 )