feat: add parquet unified schema and cleanup logging messages

Adds a unified parquet stream for all log messages with a message
matcher for necessary endpoint and connection node messages.
Logging fields were cleaned up for consistency and status code
logging uses a 0 instead of "" to avoid mixing types.

Closes #874, #882

autopush/web/webpush.py

@@ -413,11 +413,11 @@ def post(self,
         user_data = subscription["user_data"]
         self._client_info.update(
             message_id=notification.message_id,
-            uaid=hasher(user_data.get("uaid")),
+            uaid_hash=hasher(user_data.get("uaid")),
             channel_id=user_data.get("chid"),
             router_key=user_data["router_type"],
             message_size=len(notification.data or ""),
-            ttl=notification.ttl,
+            message_ttl=notification.ttl,
             version=notification.version
         )
 

autopush/websocket.py

@@ -1199,7 +1199,7 @@ def send_register_finish(self, result, endpoint, chid):
         self.sendJSON(msg)
         self.ps.metrics.increment("updates.client.register",
                                   tags=self.base_tags)
-        self.log.info(format="Register", channelID=chid,
+        self.log.info(format="Register", channel_id=chid,
                       endpoint=endpoint,
                       uaid_hash=self.ps.uaid_hash,
                       user_agent=self.ps.user_agent,
@@ -1279,7 +1279,7 @@ def ver_filter(notif):
         if found:
             msg = found[0]
             size = len(msg.data) if msg.data else 0
-            self.log.info(format="Ack", router_key="webpush", channelID=chid,
+            self.log.info(format="Ack", router_key="webpush", channel_id=chid,
                           message_id=version, message_source="direct",
                           message_size=size, uaid_hash=self.ps.uaid_hash,
                           user_agent=self.ps.user_agent, code=code,
@@ -1293,7 +1293,7 @@ def ver_filter(notif):
         if found:
             msg = found[0]
             size = len(msg.data) if msg.data else 0
-            self.log.info(format="Ack", router_key="webpush", channelID=chid,
+            self.log.info(format="Ack", router_key="webpush", channel_id=chid,
                           message_id=version, message_source="stored",
                           message_size=size, uaid_hash=self.ps.uaid_hash,
                           user_agent=self.ps.user_agent, code=code,
@@ -1348,13 +1348,13 @@ def _handle_simple_ack(self, chid, version, code):
            self.ps.direct_updates[chid] <= version:
             del self.ps.direct_updates[chid]
             self.log.info(format="Ack", router_key="simplepush",
-                          channelID=chid, message_id=version,
+                          channel_id=chid, message_id=version,
                           message_source="direct",
                           uaid_hash=self.ps.uaid_hash,
                           user_agent=self.ps.user_agent, code=code,
                           **self.ps.raw_agent)
             return
-        self.log.info(format="Ack", router_key="simplepush", channelID=chid,
+        self.log.info(format="Ack", router_key="simplepush", channel_id=chid,
                       message_id=version, message_source="stored",
                       uaid_hash=self.ps.uaid_hash,
                       user_agent=self.ps.user_agent, code=code,

parquet_schemas/README.md

@@ -0,0 +1,50 @@
+Overview
+---
+These files are appended to the hindsight configurations that produce parquet
+files in s3.
+
+Structure
+---
+These files are organized by log stream. Log streams are named
+`${app}.${type}.${source}`, where `app` is always `autopush`, `type` is
+`autopush` or `autoendpoint`, and `source` is `docker.autopush` or
+`docker.autoendpoint`, but may change based on logging configuration.
+
+For each log stream there can be a `${log_stream}.cfg` file, or a directory
+`${log_stream}/` which contains .cfg files. All cfg files must contain a
+`parquet_schema`, and are interpreted as lua. If no cfg file is specified for a
+stream, then a fallback schema is used. Files without a `.cfg` extension are
+ignore.d
+
+Providing a file for a stream indicates that there is only one schema for the
+stream, and generally only requires specifying `parquet_schema`. This field
+accesses the message after PII is scrubbed. A string to string map of mozlog
+`Fields` is provided, and all values are also copied outside the map for use as
+schema columns, with field names modified by converting to lowercase, replacing
+`.`s with `_`, and adding a prefix of `fields_`.
+
+Providing a directory for a stream indicates that the stream is going to be
+split into multiple parquet schemas, so `message_matcher` and
+`s3_path_dimensions` should be specified. These fields access the message
+before PII is scrubbed, so metadata is referenced as `Hostname`, mozlog
+metadata is referenced as `Fields[Hostname]`, and mozlog fields are accessed
+like `Fields[Fields.agent]`.
+
+The default value of `message_matcher` will match the log stream for the file,
+so extending the matcher is preferrable using the lua `..` operator for string
+concatenation. In order to keep logs that don't match specific schemas, a
+fallback cfg should be provided that negates all the other message matchers in
+the directory, and uses the default mozlog parquet schema.
+
+Date and Hour are special fields that are extracted from the mozlog `Timestamp`
+value, for use in `s3_path_dimesions`. `s3_path_dimensions` is a list of
+partition names mapped to a hindsight `read_message()` source. It is standard
+to add a partition between `log` and `date` called `type` that matches the name
+of the cfg file. The default value partitions logs by log stream (heka message
+`Type`), `Date`, and `Hour` with:
+
+    s3_path_dimensions = {
+        {name="log", source="Type"},
+        {name="date", source="Fields[Date]"},
+        {name="hour", source="Fields[Hour]"},
+    }

parquet_schemas/autopush/push.cfg

@@ -0,0 +1,46 @@
+-- vim:ft=lua
+message_matcher = message_matcher .. " && (Fields[Fields.message] == 'Successful delivery' || Fields[Fields.message] == 'Router miss, message stored.' || Fields[Fields.message] == 'Request timings'"
+message_matcher = message_matcher .. " || Fields[Fields.message] == 'hello' || Fields[Fields.message] == 'Register' || Fields[Fields.message] == 'Unregister' || Fields[Fields.message] == 'Ack' || Fields[Fields.message] == 'Nack')"
+
+s3_path_dimensions = {
+	{name="log", source="Type"},
+	{name="date", source="Fields[Date]"},
+	{name="hour", source="Fields[Hour]"},
+}
+
+parquet_schema = [=[
+message Log {
+    required int64 Timestamp;
+    optional binary Type (UTF8);
+    optional binary Hostname (UTF8);
+    optional binary Logger (UTF8);
+    optional binary EnvVersion (UTF8);
+    optional int64 Severity;
+    optional int64 Pid;
+
+    optional int64 fields_code;
+    optional binary fields_remote_ip (UTF8);
+    optional binary fields_uaid_hash (UTF8);
+    optional binary fields_message (UTF8);
+    optional boolean fields_error;
+    optional binary fields_user_agent (UTF8);
+    optional binary fields_channel_id (UTF8);
+    optional binary fields_jwt_aud (UTF8);
+    optional binary fields_jwt_crypto_key (UTF8);
+    optional binary fields_jwt_sub (UTF8);
+    optional binary fields_message_id (UTF8);
+    optional int64 fields_message_size;
+    optional binary fields_message_source (UTF8);
+    optional int64 fields_message_ttl;
+    optional int64 fields_request_time;
+    optional int64 fields_route_time;
+    optional int64 fields_validation_time;
+    optional binary fields_router_key (UTF8);
+    optional int64 fields_status_code;
+    optional binary fields_ua_browser_family (UTF8);
+    optional binary fields_ua_browser_ver (UTF8);
+    optional binary fields_ua_os_family (UTF8);
+    optional binary fields_ua_os_ver (UTF8);
+    optional binary fields_system (UTF8);
+}
+]=]