From 952d42a7e9a429a1a26d102f333c9f532e965562 Mon Sep 17 00:00:00 2001
From: Josh Durgin <jdurgin@redhat.com>
Date: Mon, 9 Apr 2018 21:47:46 -0400
Subject: [PATCH] osd/ECBackend: store the original needed shards in a ReadOp

This avoids extra logic recomputing them due to EIO or redundant
reads.

Signed-off-by: Josh Durgin <jdurgin@redhat.com>
(cherry picked from commit 0f9474d4e8e54cb8784cc2dbaa50318c2ff8e667)

Conflicts:
	src/osd/ECBackend.cc (trivial)
---
 src/osd/ECBackend.cc | 10 ++++++++++
 src/osd/ECBackend.h  |  6 +++++-
 2 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/src/osd/ECBackend.cc b/src/osd/ECBackend.cc
index 5eccb1887e723..a0fecedda32a7 100644
--- a/src/osd/ECBackend.cc
+++ b/src/osd/ECBackend.cc
@@ -254,14 +254,17 @@ struct OnRecoveryReadComplete :
 struct RecoveryMessages {
   map<hobject_t,
       ECBackend::read_request_t> reads;
+  map<hobject_t, set<int>> want_to_read;
   void read(
     ECBackend *ec,
     const hobject_t &hoid, uint64_t off, uint64_t len,
+    set<int> &&_want_to_read,
     const set<pg_shard_t> &need,
     bool attrs) {
     list<boost::tuple<uint64_t, uint64_t, uint32_t> > to_read;
     to_read.push_back(boost::make_tuple(off, len, 0));
     assert(!reads.count(hoid));
+    want_to_read.insert(make_pair(hoid, std::move(_want_to_read)));
     reads.insert(
       make_pair(
 	hoid,
@@ -526,6 +529,7 @@ void ECBackend::dispatch_recovery_messages(RecoveryMessages &m, int priority)
     return;
   start_read_op(
     priority,
+    m.want_to_read,
     m.reads,
     OpRequestRef(),
     false, true);
@@ -571,6 +575,7 @@ void ECBackend::continue_recovery_op(
 	op.hoid,
 	op.recovery_progress.data_recovered_to,
 	amount,
+	std::move(want),
 	to_read,
 	op.recovery_progress.first && !op.obc);
       op.extent_requested = make_pair(
@@ -1606,6 +1611,7 @@ int ECBackend::get_remaining_shards(
 
 void ECBackend::start_read_op(
   int priority,
+  map<hobject_t, set<int>> &want_to_read,
   map<hobject_t, read_request_t> &to_read,
   OpRequestRef _op,
   bool do_redundant_reads,
@@ -1621,6 +1627,7 @@ void ECBackend::start_read_op(
       do_redundant_reads,
       for_recovery,
       _op,
+      std::move(want_to_read),
       std::move(to_read))).first->second;
   dout(10) << __func__ << ": starting " << op << dendl;
   if (_op) {
@@ -2274,6 +2281,7 @@ void ECBackend::objects_read_and_reconstruct(
     return;
   }
 
+  map<hobject_t, set<int>> obj_want_to_read;
   set<int> want_to_read;
   get_want_to_read_shards(&want_to_read);
     
@@ -2301,10 +2309,12 @@ void ECBackend::objects_read_and_reconstruct(
 	  shards,
 	  false,
 	  c)));
+    obj_want_to_read.insert(make_pair(to_read.first, want_to_read));
   }
 
   start_read_op(
     CEPH_MSG_PRIO_DEFAULT,
+    obj_want_to_read,
     for_read_op,
     OpRequestRef(),
     fast_read, false);
diff --git a/src/osd/ECBackend.h b/src/osd/ECBackend.h
index 8799ec991fc19..7e0c2d08d47cb 100644
--- a/src/osd/ECBackend.h
+++ b/src/osd/ECBackend.h
@@ -386,6 +386,7 @@ public:
 
     ZTracer::Trace trace;
 
+    map<hobject_t, set<int>> want_to_read;
     map<hobject_t, read_request_t> to_read;
     map<hobject_t, read_result_t> complete;
 
@@ -402,9 +403,11 @@ public:
       bool do_redundant_reads,
       bool for_recovery,
       OpRequestRef op,
+      map<hobject_t, set<int>> &&_want_to_read,
       map<hobject_t, read_request_t> &&_to_read)
       : priority(priority), tid(tid), op(op), do_redundant_reads(do_redundant_reads),
-	for_recovery(for_recovery), to_read(std::move(_to_read)) {
+	for_recovery(for_recovery), want_to_read(std::move(_want_to_read)),
+	to_read(std::move(_to_read)) {
       for (auto &&hpair: to_read) {
 	auto &returned = complete[hpair.first].returned;
 	for (auto &&extent: hpair.second.to_read) {
@@ -430,6 +433,7 @@ public:
   map<pg_shard_t, set<ceph_tid_t> > shard_to_read_map;
   void start_read_op(
     int priority,
+    map<hobject_t, set<int>> &want_to_read,
     map<hobject_t, read_request_t> &to_read,
     OpRequestRef op,
     bool do_redundant_reads, bool for_recovery);
-- 
2.39.5