diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/H5FDprivate.h | 9 | ||||
| -rw-r--r-- | src/H5FDvfd_swmr.c | 10 | ||||
| -rw-r--r-- | src/H5FDvfd_swmr_private.h | 2 | ||||
| -rw-r--r-- | src/H5Fpkg.h | 52 | ||||
| -rw-r--r-- | src/H5Fpublic.h | 20 |
5 files changed, 56 insertions, 37 deletions
diff --git a/src/H5FDprivate.h b/src/H5FDprivate.h index 0d05b15..8acbd30 100644 --- a/src/H5FDprivate.h +++ b/src/H5FDprivate.h @@ -157,6 +157,15 @@ * lower file and is therefore about to be removed from the * metadata file * + * garbage: `true` if the entry is marked for garbage collection and is + * thus invalid. + * + * For n the number of entries, deleting an entry is O(n). + * H5PB_dest() deletes all entries. Instead of deleting + * entries one-by-one at O(n^2) cost, H5PB_dest() marks + * each disused entry for garbage collection and sweeps all + * entries up before it is done. + * *---------------------------------------------------------------------------- */ typedef struct H5FD_vfd_swmr_idx_entry_t { diff --git a/src/H5FDvfd_swmr.c b/src/H5FDvfd_swmr.c index e2f8513..5d2474f 100644 --- a/src/H5FDvfd_swmr.c +++ b/src/H5FDvfd_swmr.c @@ -53,13 +53,21 @@ typedef struct H5FD_vfd_swmr_t { uint32_t api_elapsed_nslots; uint64_t *api_elapsed_ticks; /* Histogram of ticks elapsed * inside the API (reader only). + * api_elapsed_ticks[elapsed] is + * the number of times `elapsed` + * ticks passed in an API call + * during the program lifetime. */ hbool_t pb_configured; /* boolean flag set to TRUE */ /* when the page buffer is */ /* and to FALSE otherwise. */ /* Used for sanity checking. */ H5F_vfd_swmr_config_t config; - bool writer; /* True iff configured to write. */ + bool writer; /* True iff configured to write. + * All methods on a write-mode + * SWMR VFD instance are passed + * to the lower VFD instance. + */ } H5FD_vfd_swmr_t; #define MAXADDR (((haddr_t)1<<(8*sizeof(HDoff_t)-1))-1) diff --git a/src/H5FDvfd_swmr_private.h b/src/H5FDvfd_swmr_private.h index 67a72ec..74a937f 100644 --- a/src/H5FDvfd_swmr_private.h +++ b/src/H5FDvfd_swmr_private.h @@ -53,7 +53,7 @@ typedef struct eot_queue_entry { hbool_t vfd_swmr_writer; uint64_t tick_num; struct timespec end_of_tick; - struct H5F_t *vfd_swmr_file; /* NOTE: for the time being use H5F_t instead H5F_file_t */ + struct H5F_t *vfd_swmr_file; /* NOTE: for the time being use H5F_t instead H5F_shared_t */ TAILQ_ENTRY(eot_queue_entry) link; } eot_queue_entry_t; diff --git a/src/H5Fpkg.h b/src/H5Fpkg.h index 1136cb4..71d0ce4 100644 --- a/src/H5Fpkg.h +++ b/src/H5Fpkg.h @@ -217,24 +217,14 @@ typedef struct H5F_mtab_t { H5F_mount_t *child; /* An array of mount records */ } H5F_mtab_t; -/* - * VFD SWMR: Entry for the delayed free space release doubly linked list - * - * md_file_page_offset: Unsigned 64-bit value containing the base address - * of the metadata page, or multi page metadata entry - * in the metadata file IN PAGES. - * To obtain byte offset, multiply this value by the - * page size. - * length: The length of the metadata page or multi page - * metadata entry in BYTES. - * tick_num: Sequence # of the current tick - * link: tailqueue linkage +/* Deferred-free record for the shadow file: records a region of bytes in + * the shadow file to release after max_lag ticks. */ typedef struct shadow_defree { - uint64_t offset; - uint32_t length; - uint64_t tick_num; - TAILQ_ENTRY(shadow_defree) link; + uint64_t offset; // offset of the region in *bytes* + uint32_t length; // length of the region in *bytes* + uint64_t tick_num; // tick number when the free was deferred + TAILQ_ENTRY(shadow_defree) link; // deferred-free queue linkage } shadow_defree_t; /* Structure specifically to store superblock. This was originally @@ -256,17 +246,27 @@ typedef struct H5F_super_t { H5G_entry_t *root_ent; /* Root group symbol table entry */ } H5F_super_t; -/* VFD SWMR: deferred free on the lower VFD. */ +/* Deferred-free record for the lower file: records a region of bytes in + * the file below the SWMR VFD to release after a delay. + */ typedef struct lower_defree { - SIMPLEQ_ENTRY(lower_defree) link; - H5FD_mem_t alloc_type; - haddr_t addr; - hsize_t size; - uint64_t free_after_tick; + SIMPLEQ_ENTRY(lower_defree) link; // deferred-free queue linkage + H5FD_mem_t alloc_type; // type with which the region was allocated + haddr_t addr; // start of the region *in bytes* + hsize_t size; // length of the region *in bytes* + uint64_t free_after_tick; /* the region may be reused on tick + * free_after_tick + 1 at the earliest + */ } lower_defree_t; +/* Queue of deferred-free records (lower_defree_t) for the lower file, sorted + * head-to-tail in increasing `free_after_tick` order. + */ typedef SIMPLEQ_HEAD(lower_defree_queue, lower_defree) lower_defree_queue_t; +/* Queue of deferred-free records (shadow_defree_t) for the shadow file, sorted + * head-to-tail in increasing `tick_num` order. + */ typedef TAILQ_HEAD(shadow_defree_queue, shadow_defree) shadow_defree_queue_t; /* @@ -403,7 +403,9 @@ struct H5F_shared_t { * configuration from the * FAPL used to open the file */ - haddr_t writer_index_offset; + haddr_t writer_index_offset; /* Current byte offset of the + * shadow index in the shadow file. + */ hbool_t vfd_swmr; /* The file is opened with VFD * SWMR configured or not */ @@ -413,7 +415,9 @@ struct H5F_shared_t { uint64_t tick_num; /* Number of the current tick */ struct timespec end_of_tick; /* End time of the current tick */ - lower_defree_queue_t lower_defrees; /* For use by VFD SWMR writers. */ + lower_defree_queue_t lower_defrees; /* Records of lower-file space + * awaiting reclamation. + */ /* VFD SWMR metadata file index */ H5FD_vfd_swmr_idx_entry_t * mdf_idx; /* pointer to an array of instance * of H5FD_vfd_swmr_idx_entry_t of diff --git a/src/H5Fpublic.h b/src/H5Fpublic.h index d4f6341..17d8ce1 100644 --- a/src/H5Fpublic.h +++ b/src/H5Fpublic.h @@ -260,17 +260,15 @@ typedef herr_t (*H5F_flush_cb_t)(hid_t object_id, void *udata); * is selected. * * md_pages_reserved: - * An integer field indicating the number of pages reserved - * at the head of the metadata file. This value must be greater than - * or equal to 1. - * When the metadata file is created, the specified number of pages is - * reserved at the head of the metadata file. In the current - * implementation, the size of the metadata file header plus the - * index is limited to this size. - * Further, in the POSIX case, when readers check for an updated index, - * this check will start with a read of md_pages_reserved pages from - * the head of the metadata file. - * + * The `md_pages_reserved` parameter tells how many pages to reserve + * at the beginning of the shadow file for the shadow-file header + * and the shadow index. The header has an entire page to itself. + * The remaining `md_pages_reserved - 1` pages are reserved for the + * shadow index. If the index grows larger than its initial + * allocation, then it will move to a new location in the shadow file, + * and the initial allocation will be reclaimed. `md_pages_reserved` + * must be at least 2. + * * pb_expansion_threshold: * An integer field indicating the threshold for the page buffer size. * During a tick, the page buffer must expand as necessary to retain copies |