Change retries to a ratelimit-based system (#35)

This is needed because both IA and CCF are now rate-limiting their CDX API endpoints.

cdx_toolkit/myrequests.py

@@ -7,22 +7,67 @@
 
 LOGGER = logging.getLogger(__name__)
 
-
 previously_seen_hostnames = {
     'commoncrawl.s3.amazonaws.com',
     'data.commoncrawl.org',
     'web.archive.org',
 }
 
 
-def dns_fatal(url):
+def dns_fatal(hostname):
     '''We have a dns error, should we fail immediately or not?'''
-    hostname = urlparse(url).hostname
     if hostname not in previously_seen_hostnames:
         return True
 
 
+retry_info = {
+    'default': {
+        'next_fetch': 0,
+        'minimum_interval': 3.0,
+    },
+    'index.commoncrawl.org': {
+        'next_fetch': 0,
+        'minimum_interval': 3.0,
+    },
+    'data.commoncrawl.org': {
+        'next_fetch': 0,
+        'minimum_interval': 3.0,
+    },
+    'web.archive.org': {
+        'next_fetch': 0,
+        'minimum_interval': 6.0,
+    },
+}
+
+
+def get_retries(hostname):
+    if hostname not in retry_info:
+        retry_info[hostname] = retry_info['default'].copy()
+        LOGGER.debug('initializing retry info for new host '+hostname)
+    entry = retry_info[hostname]
+    if not entry['next_fetch']:
+        entry['next_fetch'] = time.time()
+    return entry['next_fetch'], entry['minimum_interval']
+
+
+def update_next_fetch(hostname, next_fetch):
+    retry_info[hostname]['next_fetch'] = next_fetch
+
+
 def myrequests_get(url, params=None, headers=None, cdx=False, allow404=False):
+    t = time.time()
+
+    hostname = urlparse(url).hostname
+    next_fetch, minimum_interval = get_retries(hostname)
+
+    if t < next_fetch:
+        dt = next_fetch - t
+        if dt > 3.1:
+            LOGGER.debug('sleeping for {:.3f}s before next fetch'.format(dt))
+        time.sleep(dt)
+    # next_fetch is also updated at the bottom
+    update_next_fetch(hostname, next_fetch + minimum_interval)
+
     if params:
         if 'from_ts' in params:
             params['from'] = params['from_ts']
@@ -38,8 +83,8 @@ def myrequests_get(url, params=None, headers=None, cdx=False, allow404=False):
         headers['User-Agent'] = 'pypi_cdx_toolkit/'+__version__
 
     retry = True
-    retry_sec = 1
-    retry_max_sec = 30
+    retry_sec = 2 * minimum_interval
+    retry_max_sec = 60
     retries = 0
     connect_errors = 0
     while retry:
@@ -62,14 +107,10 @@ def myrequests_get(url, params=None, headers=None, cdx=False, allow404=False):
                 # I have never seen IA or CC send 429 or 509, but just in case...
                 # 429 is also a slow down, IA started sending them mid-2023
                 retries += 1
-                if retries > 5:
-                    LOGGER.warning('retrying after 1s for %d', resp.status_code)
-                    if resp.text:
-                        LOGGER.warning('response body is %s', resp.text)
-                else:
-                    LOGGER.info('retrying after 1s for %d', resp.status_code)
-                    if resp.text:
-                        LOGGER.info('response body is %s', resp.text)
+                level = 30 if retries > 5 else 20  # 30=warning 20=info
+                LOGGER.log(level, 'retrying after %.2fs for %d', retry_sec, resp.status_code)
+                if resp.text:
+                    LOGGER.log(level, 'response body is %s', resp.text)
                 time.sleep(retry_sec)
                 retry_sec = min(retry_sec*2, retry_max_sec)
                 continue
@@ -93,8 +134,8 @@ def myrequests_get(url, params=None, headers=None, cdx=False, allow404=False):
                 raise ValueError(string)
             if connect_errors > 10:
                 LOGGER.warning(string)
-            LOGGER.info('retrying after 1s for '+str(e))
-            time.sleep(retry_sec)
+            LOGGER.info('retrying after {:.2f}s for '.format(retry_max_sec)+str(e))
+            time.sleep(retry_max_sec)  # notice the extra-long sleep
             retry_sec = min(retry_sec*2, retry_max_sec)
         except requests.exceptions.RequestException as e:  # pragma: no cover
             LOGGER.warning('something unexpected happened, giving up after %s', str(e))
@@ -104,4 +145,7 @@ def myrequests_get(url, params=None, headers=None, cdx=False, allow404=False):
     if hostname not in previously_seen_hostnames:
         previously_seen_hostnames.add(hostname)
 
+    # in case we had a lot of retries, etc
+    update_next_fetch(hostname, time.time() + minimum_interval)
+
     return resp