Merge pull request #58787 from xxhdx1985126/wip-67106

crimson/os/seastore: consider segment_header_t::modify_time as the segments' modify_time for no-tail OOL segments

Reviewed-by: Yingxin Cheng <yingxin.cheng@intel.com>

5 files changed, 63 insertions, 14 deletions

diff --git a/src/crimson/os/seastore/async_cleaner.cc b/src/crimson/os/seastore/async_cleaner.cc
index 7e13befde0d..5046980eae5 100644
--- a/src/crimson/os/seastore/async_cleaner.cc
+++ b/src/crimson/os/seastore/async_cleaner.cc
@@ -1372,20 +1372,33 @@ SegmentCleaner::mount_ret SegmentCleaner::mount()
         segment_id
       ).safe_then([this, FNAME, segment_id, header](auto tail)
         -> scan_extents_ertr::future<> {
-        if (tail.segment_nonce != header.segment_nonce) {
+	bool tail_valid = (tail.segment_nonce == header.segment_nonce);
+        if (!tail_valid && header.type == segment_type_t::JOURNAL) {
           return scan_no_tail_segment(header, segment_id);
         }
-        ceph_assert(header.get_type() == tail.get_type());
-
-        sea_time_point modify_time = mod_to_timepoint(tail.modify_time);
-        std::size_t num_extents = tail.num_extents;
-        if ((modify_time == NULL_TIME && num_extents == 0) ||
-            (modify_time != NULL_TIME && num_extents != 0)) {
-          segments.update_modify_time(segment_id, modify_time, num_extents);
-        } else {
-          ERROR("illegal modify time {}", tail);
-          return crimson::ct_error::input_output_error::make();
-        }
+        ceph_assert(header.get_type() == tail.get_type() || !tail_valid);
+
+	sea_time_point header_time = mod_to_timepoint(header.modify_time);
+	if (tail_valid) {
+	  sea_time_point tail_time = mod_to_timepoint(tail.modify_time);
+	  std::size_t num_extents = tail.num_extents;
+	  DEBUG("updating modify time for segment {}, "
+		"mod time {}-{}, num_extents {}",
+		segment_id, header_time, tail_time, num_extents);
+	  if (num_extents == 0) {
+	    ceph_assert(tail_time == NULL_TIME);
+	  } else {
+	    ceph_assert(header_time != NULL_TIME);
+	    ceph_assert(tail_time != NULL_TIME);
+	    sea_time_point avg_time = get_average_time(
+	      header_time, 1, tail_time, 1);
+	    segments.update_modify_time(segment_id, avg_time, num_extents);
+	  }
+	} else {
+	  DEBUG("updating modify time for segment {}, mod time {}, without tail",
+		segment_id, header_time);
+	  segments.init_modify_time(segment_id, header_time);
+	}
 
         init_mark_segment_closed(
           segment_id,
@@ -1396,8 +1409,22 @@ SegmentCleaner::mount_ret SegmentCleaner::mount()
         return seastar::now();
       }).handle_error(
         crimson::ct_error::enodata::handle(
-          [this, header, segment_id](auto) {
-          return scan_no_tail_segment(header, segment_id);
+          [this, header, segment_id, FNAME](auto) {
+	  if (header.type == segment_type_t::JOURNAL) {
+	    return scan_no_tail_segment(header, segment_id);
+	  } else {
+	    sea_time_point modify_time = mod_to_timepoint(header.modify_time);
+	    DEBUG("updating modify time for segment {}, mod time {}",
+	      segment_id, modify_time);
+	    segments.init_modify_time(segment_id, modify_time);
+	    init_mark_segment_closed(
+	      segment_id,
+	      header.segment_seq,
+	      header.type,
+	      header.category,
+	      header.generation);
+	    return scan_extents_ertr::now();
+	  }
         }),
         crimson::ct_error::pass_further_all{}
       );
diff --git a/src/crimson/os/seastore/async_cleaner.h b/src/crimson/os/seastore/async_cleaner.h
index a90d02e4fcd..424247c5bdc 100644
--- a/src/crimson/os/seastore/async_cleaner.h
+++ b/src/crimson/os/seastore/async_cleaner.h
@@ -43,6 +43,7 @@ struct segment_info_t {
 
   sea_time_point modify_time = NULL_TIME;
 
+  // Might be unavailable(0), see mount() -> init_modify_time()
   std::size_t num_extents = 0;
 
   segment_off_t written_to = 0;
@@ -75,6 +76,13 @@ struct segment_info_t {
 
   void set_closed();
 
+  void init_modify_time(sea_time_point _modify_time) {
+    ceph_assert(modify_time == NULL_TIME);
+    ceph_assert(num_extents == 0);
+    ceph_assert(_modify_time != NULL_TIME);
+    modify_time = _modify_time;
+  }
+
   void update_modify_time(sea_time_point _modify_time, std::size_t _num_extents) {
     ceph_assert(!is_closed());
     assert(_modify_time != NULL_TIME);
@@ -226,6 +234,15 @@ public:
 
   void update_written_to(segment_type_t, paddr_t);
 
+  void init_modify_time(
+      segment_id_t id, sea_time_point tp) {
+    if (tp == NULL_TIME) {
+      return;
+    }
+
+    segments[id].init_modify_time(tp);
+  }
+
   void update_modify_time(
       segment_id_t id, sea_time_point tp, std::size_t num) {
     if (num == 0) {
diff --git a/src/crimson/os/seastore/journal/segment_allocator.cc b/src/crimson/os/seastore/journal/segment_allocator.cc
index 61e1be585c8..7760c869b49 100644
--- a/src/crimson/os/seastore/journal/segment_allocator.cc
+++ b/src/crimson/os/seastore/journal/segment_allocator.cc
@@ -93,6 +93,7 @@ SegmentAllocator::do_open(bool is_mkfs)
       alloc_tail = JOURNAL_SEQ_NULL;
     }
     auto header = segment_header_t{
+      timepoint_to_mod(seastar::lowres_system_clock::now()),
       new_segment_seq,
       segment_id,
       dirty_tail,
diff --git a/src/crimson/os/seastore/seastore_types.cc b/src/crimson/os/seastore/seastore_types.cc
index 25d787f0fd3..5481a680f27 100644
--- a/src/crimson/os/seastore/seastore_types.cc
+++ b/src/crimson/os/seastore/seastore_types.cc
@@ -371,6 +371,7 @@ std::ostream &operator<<(std::ostream &out, const segment_header_t &header)
              << ", dirty_tail=" << header.dirty_tail
              << ", alloc_tail=" << header.alloc_tail
              << ", segment_nonce=" << header.segment_nonce
+	     << ", modify_time=" << mod_time_point_printer_t{header.modify_time}
              << ")";
 }
 
diff --git a/src/crimson/os/seastore/seastore_types.h b/src/crimson/os/seastore/seastore_types.h
index dc2f478136e..35c331d5096 100644
--- a/src/crimson/os/seastore/seastore_types.h
+++ b/src/crimson/os/seastore/seastore_types.h
@@ -1799,6 +1799,8 @@ using segment_nonce_t = uint32_t;
  * 4) Replay from the latest tails
  */
 struct segment_header_t {
+  mod_time_point_t modify_time;
+
   segment_seq_t segment_seq;
   segment_id_t physical_segment_id; // debugging
 
@@ -1817,6 +1819,7 @@ struct segment_header_t {
 
   DENC(segment_header_t, v, p) {
     DENC_START(1, 1, p);
+    denc(v.modify_time, p);
     denc(v.segment_seq, p);
     denc(v.physical_segment_id, p);
     denc(v.dirty_tail, p);