Ceph OSD處理客戶端寫操作處理流程的示例分析

這篇文章主要介紹Ceph OSD處理客戶端寫操作處理流程的示例分析，文中介紹的非常詳細，具有一定的參考價值，感興趣的小伙伴們一定要看完！

1、OSD從client端收到請求的處理流程

OSD::ms_fast_dispatch()

|__OSD::dispatch_session_waiting()

     |__OSD::dispatch_op_fast()

          |__OSD::handle_op()

               |__OSD::get_pg_or_queue_for_pg()

               |__OSD::enqueue_op()

                    |__PG::queue_op()

                         |__OSD::op_wq.queue()          將client的請求寫入到隊列

主OSD處理從client端請求的處理流程

OSD::ShardOpWQ::_process()

|__OSD::dequeue_op()

     |__ReplicatedPG::do_request()

          |__ReplicatedPG::do_op()

               |__ReplicatedPG::execute_ctx()

                    |__ReplicatedPG::preapare_transcation()

                         |__ReplicatedPG::do_osd_ops()

                    |__ReplicatedPG::issue_repop()

                         |__ReplicateBackend::submit_transaction()

                              |__ReplicateBackend::issue_op()                    主OSD將寫請求發送到從OSDs

                              |__ReplicatedPG::queue_transcations()          寫主OSD FileJournal

                                   |__ObjectStore::queue_transactions()

                                        |__FileStore::queue_transactions()

                                             |__JournalingObjectStore::_op_journal_transactions()

                                                  |__FileJournal::submit_entry()     將日志寫入到FileJournal的writeq隊列

                    |__ReplicatedPG::eval_repop()

主OSD的FileJournal寫線程

FileJournal::write_thread_entry()           負責從writeq隊列中獲取待寫入的日志

|__FileJournal::prepare_multi_write()     構成日志

|__FileJournal::do_write()                       實際寫日志

     |__FIleJournal::queue_completions_thru()

          |__completion_peek_front()          獲取C_JournalAhead類實例

          |__finisher->queue()

          |__finisher_cond.Signal()               啟動FileJournal的finisher線程處理C_JournalAhead類實例

主OSD的C_JournalAhead類處理

C_JournalAhead::finish()

|__FileStore::_journaled_ahead()

     |__FileStore::queue_op()

          |__FileStore::op_wq.queue()          將數據寫入到FileStore寫線程隊列中

     |__ondisk_finisher.queue(ondisk)     調用ReplicatedBackend->op_commit()回調函數

主OSD的FileStore寫數據線程類處理

FileStore::op_wq

|__FileStore::op_wq._process()

     |__FileStore::_do_op()

          |__FileStore::_do_transactions()

               |__FileStore::_do_transaction()

                    |__FileStore::_write()          落盤到FileStore中（寫入指定的文件中）

|__FileStore::op_wq._process_finish()

     |__FileStore::_finish_op()

          |__op_finisher.queue()                    調用ReplicatedPG::op_applied()回調函數


2、從OSD處理主OSD端發送過來的請求

OSD::ms_fast_dispatch()

|__OSD::dispatch_session_waiting()

     |__OSD::dispatch_op_fast()

          |__OSD::handle_replica_op()

               |__OSD::get_pg_or_queue_for_pg()

               |__OSD::enqueue_op()

                    |__PG::queue_op()

                         |__OSD::op_wq.queue()          將主OSD的請求寫入到從OSD處理隊列中

從OSD處理workqueue

OSD::ShardOpWQ::_process()

|__OSD::dequeue_op()

     |__ReplicatedPG::do_request()

          |__ReplicatedBackend::handle_message()

               |__ReplicatedBackend::sub_op_modify()

                    |__ReplicatedBackend::sub_op_modify_impl()

                         |__注冊commit回調函數類C_OSD_RepModifyCommit()

                         |__注冊Apply回調函數類C_OSD_RepModifyApply()

                         |__ReplicatedPG::queue_transcations()          從OSD寫FileJournal

                                   |__ObjectStore::queue_transactions()

                                        |__FileStore::queue_transactions()

                                             |__JournalingObjectStore::_op_journal_transactions()

                                                  |__FileJournal::submit_entry()     將日志寫入到FileJournal的writeq隊列

                    |__ReplicatedPG::eval_repop()

從OSD的FileJournal寫線程

FileJournal::write_thread_entry()           負責從writeq隊列中獲取待寫入的日志

|__FileJournal::prepare_multi_write()     構成日志

|__FileJournal::do_write()                       實際寫日志

     |__FIleJournal::queue_completions_thru()

          |__completion_peek_front()          獲取C_JournalAhead類實例

          |__finisher->queue()

          |__finisher_cond.Signal()               啟動FileJournal的finisher線程處理C_JournalAhead類實例

從OSD的C_JournalAhead類處理

C_JournalAhead::finish()

|__FileStore::_journaled_ahead()

     |__FileStore::queue_op()

          |__FileStore::op_wq.queue()          將數據寫入到FileStore寫線程隊列中

     |__ondisk_finisher.queue(ondisk)     調用ReplicatedBackend->op_commit()回調函數

從OSD的日志寫完成后的處理

C_OSD_RepModifyCommit()

|__finish()

     |__ReplicatedBackend::sub_op_modify_commit()

          |__生成MOSDRepOpReply消息                                   CEPH_OSD_FLAG_ONDISK

          |__ReplicatedPG::send_message_osd_cluster()          將從OSD回復信息發給主OSD

               |__OSD::send_message_osd_cluster()

從OSD的FileStore寫數據線程類處理

FileStore::op_wq

|__FileStore::op_wq._process()

     |__FileStore::_do_op()

          |__FileStore::_do_transactions()

               |__FileStore::_do_transaction()

                    |__FileStore::_write()          落盤到FileStore中（寫入指定的文件中）

|__FileStore::op_wq._process_finish()

     |__FileStore::_finish_op()

          |__op_finisher.queue()                    調用ReplicatedPG::op_applied()回調函數

從OSD的落盤完成后的處理

C_OSD_RepModifyApply()

|__finish()

     |__ReplicatedBackend::sub_op_modify_applied()

          |__生成MOSDRepOpReply消息                                   CEPH_OSD_FLAG_ACK

          |__ReplicatedPG::send_message_osd_cluster()          將從OSD回復信息發給主OSD

               |__OSD::send_message_osd_cluster()


3、主OSD處理從OSD發送Reply處理流程

從OSD處理主OSD端發送過來的請求

OSD::ms_fast_dispatch()

|__OSD::dispatch_session_waiting()

     |__OSD::dispatch_op_fast()

          |__OSD::handle_replica_op()

               |__OSD::get_pg_or_queue_for_pg()

               |__OSD::enqueue_op()

                    |__PG::queue_op()

                         |__OSD::op_wq.queue()          將從OSD的reply寫入到主OSD處理隊列中

主OSD處理從OSD端發送過來的reply

OSD::ShardOpWQ::_process()

|__OSD::dequeue_op()

     |__ReplicatedPG::do_request()

          |__ReplicatedBackend::handle_message()

               |__ReplicatedBackend::sub_op_modify_reply()

                    |__ip_op.waiting_for_commit.erase()

                    |__ip_op.waiting_for_applied.erase()

                    |__對于ip_op.waiting_for_applied為空，則調用ip_op.on_applied->complete(0)，即：調用on_all_applied回調函數處理。on_all_applied回調函數在ReplicatedPG.cc::execute_ctx()函數中注冊，用來對所有副本都落盤完成后的處理。on_all_applied()函數創建一個MOSDOpReply消息并且設置消息的flags=CEPH_OSD_FLAG_ACK且將該消息發送給客戶端；

                    |__對于ip_op.waiting_for_commit為空，則調用ip_op.on_commit->complete(0)，即：調用on_all_commit回調函數處理。on_all_commit回調函數在ReplicatedPG.cc::execute_ctx()函數中注冊，用來對所有副本在寫FileJournal完成后的處理。on_all_commit()函數創建一個MOSDOpReply消息并且設置消息的flags＝CEPH_OSD_FLAG_ACK|CEPH_OSD_FLAG_ONDISK且將該消息發送給客戶端；

以上是“Ceph OSD處理客戶端寫操作處理流程的示例分析”這篇文章的所有內容，感謝各位的閱讀！希望分享的內容對大家有幫助，更多相關知識，歡迎關注億速云行業資訊頻道！