Merge pull request ceph#62556 from aainscow/ec_pr_and_prereqs

osd: Optimised EC

Reviewed-by: Radoslaw Zarzynski <[email protected]>
Reviewed-by: Laura Flores <[email protected]>

Author: rzarzynski

src/common/bitset_set.h

@@ -260,8 +260,9 @@ class bitset_set {
 
   /** @return true if the container contains Key k. */
   bool contains(KeyT k) const {
-    ceph_assert(unsigned_cast(k) < max_bits);
-    ceph_assert(int(k) >= 0);
+    if (unsigned_cast(k) >= max_bits) {
+      return false;
+    }
     return (words[int(k) / bits_per_uint64_t]
       & 1ULL << (int(k) % bits_per_uint64_t));
   }

src/common/interval_map.h

@@ -209,8 +209,8 @@ class interval_map {
     }
     return ret;
   }
-  template<bool strict = true>
-  void to_interval_set(interval_set<K, C, strict> &set) const {
+  template<template<typename, typename, typename ...> class ISC = std::map, bool strict = true>
+  void to_interval_set(interval_set<K, ISC, strict> &set) const {
     for (auto &&i: *this) {
       set.insert(i.get_off(), i.get_len());
     }

src/common/options/global.yaml.in

@@ -2703,6 +2703,15 @@ options:
   services:
   - mon
   with_legacy: true
+- name: osd_pool_default_flag_ec_optimizations
+  type: bool
+  level: advanced
+  desc: Control whether to create new erasure coded pools with EC optimizations turned on by default.
+  fmt_desc: Set the ``allow_ec_optimizations`` flag on new erasure coded pools.
+  default: false
+  services:
+  - mon
+  with_legacy: true
 - name: osd_pool_default_hit_set_bloom_fpp
   type: float
   level: advanced
@@ -6765,3 +6774,15 @@ options:
     The format is ``{file}:{line} [, {file}:{line}]``
   level: dev
   with_legacy: false
+- name: ec_extent_cache_size
+  type: uint
+  level: advanced
+  desc: Size of per-shard extent cache
+  default: 10485760
+  services:
+  - osd
+- name: ec_pdw_write_mode
+  type: uint
+  level: dev
+  default: 0
+  desc: When EC writes should generate PDWs (development only) 0=optimal 1=never 2=when possible

src/crimson/osd/pg.h

@@ -478,6 +478,10 @@ class PG : public boost::intrusive_ref_counter<
     void trim(const pg_log_entry_t &entry) override {
       // TODO
     }
+    void partial_write(pg_info_t *info, const pg_log_entry_t &entry) override {
+      // TODO
+      ceph_assert(entry.written_shards.empty() && info->partial_writes_last_complete.empty());
+    }
   };
   PGLog::LogEntryHandlerRef get_log_handler(
     ceph::os::Transaction &t) final {

src/crimson/osd/scrub/pg_scrubber.cc

@@ -145,7 +145,6 @@ chunk_validation_policy_t PGScrubber::get_policy() const
 {
   return chunk_validation_policy_t{
     pg.get_primary(),
-    std::nullopt /* stripe_info, populate when EC is implemented */,
     crimson::common::local_conf().get_val<Option::size_t>(
       "osd_max_object_size"),
     crimson::common::local_conf().get_val<std::string>(

src/crimson/osd/scrub/scrub_validator.h

@@ -9,14 +9,12 @@
 #include "common/config_proxy.h"
 #include "common/scrub_types.h"
 #include "crimson/common/log.h"
-#include "osd/ECUtil.h"
 #include "osd/osd_types.h"
 
 namespace crimson::osd::scrub {
 
 struct chunk_validation_policy_t {
   pg_shard_t primary;
-  std::optional<ECUtil::stripe_info_t> stripe_info;
 
   // osd_max_object_size
   size_t max_object_size;
@@ -31,11 +29,13 @@ struct chunk_validation_policy_t {
 
 
   bool is_ec() const {
-    return !!stripe_info;
+    // FIXME: See scrub_backend in classic for reference.
+    return false;
   }
 
   size_t logical_to_ondisk_size(size_t size) const {
-    return stripe_info ? stripe_info->logical_to_next_chunk_offset(size) : size;
+    // FIXME: See scrub_backend in classic for how to handle EC.
+    return size;
   }
 };
 

src/erasure-code/ErasureCodeInterface.h

@@ -677,6 +677,10 @@ namespace ceph {
        * clay). Other plugins will not process the overhead of stub sub-chunks.
        */
       FLAG_EC_PLUGIN_REQUIRE_SUB_CHUNKS = 1<<5,
+      /* Optimized EC is supported only if this flag is set. All other flags
+       * are irrelevant if this flag is false.
+       */
+      FLAG_EC_PLUGIN_OPTIMIZED_SUPPORTED = 1<<6,
     };
     static const char *get_optimization_flag_name(const plugin_flags flag) {
       switch (flag) {
@@ -686,6 +690,7 @@ namespace ceph {
       case FLAG_EC_PLUGIN_ZERO_PADDING_OPTIMIZATION: return "zeropadding";
       case FLAG_EC_PLUGIN_PARITY_DELTA_OPTIMIZATION: return "paritydelta";
       case FLAG_EC_PLUGIN_REQUIRE_SUB_CHUNKS: return "requiresubchunks";
+      case FLAG_EC_PLUGIN_OPTIMIZED_SUPPORTED: return "optimizedsupport";
       default: return "???";
       }
     }

src/erasure-code/isa/ErasureCodeIsa.h

@@ -26,10 +26,13 @@
 #define CEPH_ERASURE_CODE_ISA_L_H
 
 // -----------------------------------------------------------------------------
+#include <string_view>
 #include "erasure-code/ErasureCode.h"
 #include "ErasureCodeIsaTableCache.h"
 // -----------------------------------------------------------------------------
 
+using namespace std::literals;
+
 #define EC_ISA_ADDRESS_ALIGNMENT 32u
 
 #define is_aligned(POINTER, BYTE_COUNT) \
@@ -51,6 +54,7 @@ class ErasureCodeIsa : public ceph::ErasureCode {
 
   ErasureCodeIsaTableCache &tcache;
   const char *technique;
+  uint64_t flags;
 
   ErasureCodeIsa(const char *_technique,
                  ErasureCodeIsaTableCache &_tcache) :
@@ -60,6 +64,15 @@ class ErasureCodeIsa : public ceph::ErasureCode {
   tcache(_tcache),
   technique(_technique)
   {
+    flags = FLAG_EC_PLUGIN_PARTIAL_READ_OPTIMIZATION |
+            FLAG_EC_PLUGIN_PARTIAL_WRITE_OPTIMIZATION |
+            FLAG_EC_PLUGIN_ZERO_INPUT_ZERO_OUTPUT_OPTIMIZATION |
+            FLAG_EC_PLUGIN_PARITY_DELTA_OPTIMIZATION;
+
+    if (technique == "reed_sol_van"sv ||
+        technique == "default"sv) {
+      flags |= FLAG_EC_PLUGIN_OPTIMIZED_SUPPORTED;
+    }
   }
 
 
@@ -68,10 +81,7 @@ class ErasureCodeIsa : public ceph::ErasureCode {
   }
 
   uint64_t get_supported_optimizations() const override {
-    return FLAG_EC_PLUGIN_PARTIAL_READ_OPTIMIZATION |
-      FLAG_EC_PLUGIN_PARTIAL_WRITE_OPTIMIZATION |
-      FLAG_EC_PLUGIN_ZERO_INPUT_ZERO_OUTPUT_OPTIMIZATION |
-      FLAG_EC_PLUGIN_PARITY_DELTA_OPTIMIZATION;
+    return flags;
   }
 
   unsigned int

src/erasure-code/jerasure/ErasureCodeJerasure.h

@@ -18,8 +18,12 @@
 #ifndef CEPH_ERASURE_CODE_JERASURE_H
 #define CEPH_ERASURE_CODE_JERASURE_H
 
+#include <string_view>
+
 #include "erasure-code/ErasureCode.h"
 
+using namespace std::literals;
+
 class ErasureCodeJerasure : public ceph::ErasureCode {
 public:
   int k;
@@ -32,28 +36,33 @@ class ErasureCodeJerasure : public ceph::ErasureCode {
   std::string rule_root;
   std::string rule_failure_domain;
   bool per_chunk_alignment;
+  uint64_t flags;
+
+  explicit ErasureCodeJerasure(const char *_technique)
+      : k(0),
+        DEFAULT_K("2"),
+        m(0),
+        DEFAULT_M("1"),
+        w(0),
+        DEFAULT_W("8"),
+        technique(_technique),
+        per_chunk_alignment(false) {
+    flags = FLAG_EC_PLUGIN_PARTIAL_READ_OPTIMIZATION |
+      FLAG_EC_PLUGIN_PARTIAL_WRITE_OPTIMIZATION |
+      FLAG_EC_PLUGIN_ZERO_INPUT_ZERO_OUTPUT_OPTIMIZATION |
+      FLAG_EC_PLUGIN_PARITY_DELTA_OPTIMIZATION;
 
-  explicit ErasureCodeJerasure(const char *_technique) :
-    k(0),
-    DEFAULT_K("2"),
-    m(0),
-    DEFAULT_M("1"),
-    w(0),
-    DEFAULT_W("8"),
-    technique(_technique),
-    per_chunk_alignment(false)
-  {}
+    if (technique == "reed_sol_van"sv) {
+      flags |= FLAG_EC_PLUGIN_OPTIMIZED_SUPPORTED;
+    }
+  }
 
   ~ErasureCodeJerasure() override {}
 
   uint64_t get_supported_optimizations() const override {
-    return FLAG_EC_PLUGIN_PARTIAL_READ_OPTIMIZATION |
-      FLAG_EC_PLUGIN_PARTIAL_WRITE_OPTIMIZATION |
-      FLAG_EC_PLUGIN_ZERO_INPUT_ZERO_OUTPUT_OPTIMIZATION |
-      FLAG_EC_PLUGIN_PARITY_DELTA_OPTIMIZATION;
+    return flags;
   }
 
-  
   unsigned int get_chunk_count() const override {
     return k + m;
   }

src/include/mempool.h

@@ -178,6 +178,7 @@ namespace mempool {
   f(osdmap_mapping)		      \
   f(pgmap)			      \
   f(mds_co)			      \
+  f(ec_extent_cache)                  \
   f(unittest_1)			      \
   f(unittest_2)