LogDog: Update to archival V2.

common/proto/logdog/svcconfig/config.proto

 // Copyright 2015 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 syntax = "proto3";
 
 package svcconfig;
 
 import "transport.proto";
 import "storage.proto";
@@ skipping 23 matching lines @@
   // The name of the authentication group for administrators.
   string admin_auth_group = 10;
   // The name of the authentication group for backend services.
   string service_auth_group = 11;
 
   // A list of origin URLs that are allowed to perform CORS RPC calls.
   repeated string rpc_allow_origins = 20;
 
   // The name of the archive task queue.
   string archive_task_queue = 30;
-  // The amount of time after a log has been terminated before it is candidate
+
+  // The grace period after a log has been terminated and before it is candidate
   // for archival.
   //
-  // Archival triggered by this delay will NOT succeed if any log entries are
-  // missing from intermediate storage.
-  //
-  // This should be based on a period of time where it's reasonable to expect
-  // that all log messages in the transport have arrived for a given log stream.
-  // Since the transport doesn't have to guarantee in-order delivery, this
-  // should allow for the case where the terminal log entry arrives before some
-  // of the intermediate log entries. This will help avoid triggering
-  // archive attempts that are doomed to fail because of standard transport lag.
-  google.protobuf.Duration archive_delay = 31;
+  // This grace period is purely an optimization, providing time for
+  // out-of-order logs to settle and be ingested into intermediate storage and
+  // avoiding wasted Archivist cycles.
+  google.protobuf.Duration archive_settle_delay = 31;
+
   // The amount of time before a log stream is candidate for archival regardless
   // of whether or not it's been terminated or complete.
   //
-  // This endpoint is a failsafe designed to ensure that log streams with
-  // missing records or no terminal record (e.g., Butler crashed) are eventually
-  // moved out of intermediate storage.
+  // This is a failsafe designed to ensure that log streams with missing records
+  // or no terminal record (e.g., Butler crashed) are eventually archived.
   //
-  // This must be >= `archive_delay`, and should be fairly large (days) to allow
-  // for the log stream to complete and for all available log entries to be
-  // added to intermediate storage.
+  // This should be fairly large (days) to avoid prematurely archiving
+  // long-running streams, but should be considerably smaller than the
+  // intermediate storage data retention period.
   google.protobuf.Duration archive_delay_max = 32;
+
+  // The maximum number of times to retry a failed stream archival before giving
+  // up and clearing the stream.
+  //
+  // This only affects fatal archival failures (e.g., data corruption), not
+  // partial log stream content or transient failures.
+  int32 archive_retries = 33;
 }
 
 // Collector is the set of configuration parameters for Collector instances.
 message Collector {
   // The maximum number of concurrent transport messages to process. If <= 0,
   // a default will be chosen based on the transport.
   int32 max_concurrent_messages = 1;
 
   // The maximum number of concurrent workers to process each ingested message.
   // If <= 0, collector.DefaultMaxMessageWorkers will be used.
@@ skipping 20 matching lines @@
   // The bucket name must be included (e.g., "gs://foo"). The remainder of the
   // base path is optional based on desired archive location.
   string gs_base = 10;
   // If not zero, the maximum number of stream indices between index entries.
   int32 stream_index_range = 11;
   // If not zero, the maximum number of prefix indices between index entries.
   int32 prefix_index_range = 12;
   // If not zero, the maximum number of log data bytes between index entries.
   int32 byte_range = 13;
 }