[couchdb] branch prototype/fdb-layer updated: Retry for failed indexes builds

[ga...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

garren pushed a commit to branch prototype/fdb-layer
in repository https://gitbox.apache.org/repos/asf/couchdb.git


The following commit(s) were added to refs/heads/prototype/fdb-layer by this push:
     new 2247c80  Retry for failed indexes builds
2247c80 is described below

commit 2247c80397205cdaee46d4ae059690673d8696b9
Author: Garren Smith <ga...@gmail.com>
AuthorDate: Mon Nov 25 15:20:29 2019 +0200

    Retry for failed indexes builds
    
    Retry building a failing index for a set number of retries. If it
    never completes, then return the error to the user.
---
 src/couch_views/src/couch_views_indexer.erl | 74 +++++++++++++++++++++++++++--
 src/couch_views/src/couch_views_jobs.erl    |  3 +-
 test/elixir/test/map_test.exs               | 32 +++++++++++++
 3 files changed, 103 insertions(+), 6 deletions(-)

diff --git a/src/couch_views/src/couch_views_indexer.erl b/src/couch_views/src/couch_views_indexer.erl
index 7c05c1d..75e4b36 100644
--- a/src/couch_views/src/couch_views_indexer.erl
+++ b/src/couch_views/src/couch_views_indexer.erl
@@ -35,11 +35,13 @@ spawn_link() ->
 
 
 init() ->
-    {ok, Job, Data} = couch_jobs:accept(?INDEX_JOB_TYPE, #{}),
+    {ok, Job, Data0} = couch_jobs:accept(?INDEX_JOB_TYPE, #{}),
+    Data = upgrade_data(Data0),
     #{
         <<"db_name">> := DbName,
         <<"ddoc_id">> := DDocId,
-        <<"sig">> := JobSig
+        <<"sig">> := JobSig,
+        <<"retries">> := Retries
     } = Data,
 
     {ok, Db} = try
@@ -87,7 +89,63 @@ init() ->
         design_opts => Mrst#mrst.design_opts
     },
 
-    update(Db, Mrst, State).
+    try
+        update(Db, Mrst, State)
+    catch
+        exit:normal ->
+            ok;
+        Error:Reason  ->
+            NewRetry = Retries + 1,
+            RetryLimit = retry_limit(),
+
+            case should_retry(NewRetry, RetryLimit, Reason) of
+                true ->
+                    DataErr = Data#{<<"retries">> := NewRetry},
+                    % Set the last_seq to 0 so that it doesn't trigger a
+                    % successful view build for anyone listening to the
+                    % couch_views_jobs:wait_for_job
+                    % Note this won't cause the view to rebuild from 0 again
+                    StateErr = State#{job_data := DataErr, last_seq := <<"0">>},
+                    report_progress(StateErr, update);
+                false ->
+                    NewData = add_error(Error, Reason, Data),
+                    couch_jobs:finish(undefined, Job, NewData),
+                    exit(normal)
+            end
+    end.
+
+
+upgrade_data(Data) ->
+    case maps:is_key(<<"retries">>, Data) of
+        true -> Data;
+        false -> Data#{<<"retries">> =>0}
+    end.
+
+
+% Transaction limit exceeded don't retry
+should_retry(_, _, {erlfdb_error, 2101}) ->
+    false;
+
+should_retry(Retries, RetryLimit, _) when Retries < RetryLimit ->
+    true;
+
+should_retry(_, _, _) ->
+    false.
+
+
+add_error(error, {erlfdb_error, Code}, Data) ->
+    CodeBin = couch_util:to_binary(Code),
+    CodeString = erlfdb:get_error_string(Code),
+    Data#{
+        error => foundationdb_error,
+        reason => list_to_binary([CodeBin, <<"-">>, CodeString])
+    };
+
+add_error(Error, Reason, Data) ->
+    Data#{
+        error => couch_util:to_binary(Error),
+        reason => couch_util:to_binary(Reason)
+    }.
 
 
 update(#{} = Db, Mrst0, State0) ->
@@ -322,7 +380,8 @@ report_progress(State, UpdateType) ->
     #{
         <<"db_name">> := DbName,
         <<"ddoc_id">> := DDocId,
-        <<"sig">> := Sig
+        <<"sig">> := Sig,
+        <<"retries">> := Retries
     } = JobData,
 
     % Reconstruct from scratch to remove any
@@ -331,7 +390,8 @@ report_progress(State, UpdateType) ->
         <<"db_name">> => DbName,
         <<"ddoc_id">> => DDocId,
         <<"sig">> => Sig,
-        <<"view_seq">> => LastSeq
+        <<"view_seq">> => LastSeq,
+        <<"retries">> => Retries
     },
 
     case UpdateType of
@@ -356,3 +416,7 @@ report_progress(State, UpdateType) ->
 
 num_changes() ->
     config:get_integer("couch_views", "change_limit", 100).
+
+
+retry_limit() ->
+    config:get_integer("couch_views", "retry_limit", 3).
diff --git a/src/couch_views/src/couch_views_jobs.erl b/src/couch_views/src/couch_views_jobs.erl
index 87e4fea..7e0ac97 100644
--- a/src/couch_views/src/couch_views_jobs.erl
+++ b/src/couch_views/src/couch_views_jobs.erl
@@ -96,7 +96,8 @@ job_data(Db, Mrst) ->
     #{
         db_name => fabric2_db:name(Db),
         ddoc_id => DDocId,
-        sig => fabric2_util:to_hex(Sig)
+        sig => fabric2_util:to_hex(Sig),
+        retries => 0
     }.
 
 
diff --git a/test/elixir/test/map_test.exs b/test/elixir/test/map_test.exs
index fa17587..bccd417 100644
--- a/test/elixir/test/map_test.exs
+++ b/test/elixir/test/map_test.exs
@@ -503,6 +503,38 @@ defmodule ViewMapTest do
     assert keys == ["bar"]
   end
 
+  test "send error for failed indexing", context do
+    db_name = context[:db_name]
+
+    docs = [
+      %{_id: "doc1", foo: "foo", bar: "bar"},
+      %{
+        _id: "_design/view1",
+        views: %{
+          view: %{
+            map: """
+                function (doc) {
+                  for (var i=0; i<10000; i++) {
+                  emit({doc: doc._id + 1}, doc._id);
+                }
+              }
+            """
+          }
+        }
+      }
+    ]
+
+    resp = Couch.post("/#{db_name}/_bulk_docs", body: %{:docs => docs})
+    assert resp.status_code == 201
+
+    url = "/#{db_name}/_design/view1/_view/view"
+
+    resp = Couch.get(url, timeout: 500_000)
+    assert resp.status_code == 500
+    %{:body => %{"error" => error}} = resp
+    assert error == "foundationdb_error"
+  end
+
   def update_doc_value(db_name, id, value) do
     resp = Couch.get("/#{db_name}/#{id}")
     doc = convert(resp.body)