Recent changes (master)

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



The following changes since commit 72f397487788fc0b542870de5cc29ea8e8134346:

  Make options mask a 64-bit type (2015-12-21 16:03:01 -0700)

are available in the git repository at:

  git://git.kernel.dk/fio.git master

for you to fetch changes up to 5c74fc767600f75cd6d53bdc5926962b75a614ae:

  backend: continue on failed wait_for() check, not break (2015-12-22 13:40:14 -0700)

----------------------------------------------------------------
Andrey Kuzmin (1):
      Add support for waiting for other jobs by name

Fabrice Bacchella (1):
      rewrote libhdfs engine

Jens Axboe (3):
      parse: ensure that option group/category hold 64-bit
      gfio: update for 64-bit option category/group
      backend: continue on failed wait_for() check, not break

 HOWTO                |  31 ++--
 backend.c            |  32 ++++
 cconv.c              |   3 +
 engines/libhdfs.c    | 431 ++++++++++++++++++++++++++++++++++++---------------
 examples/waitfor.fio |  35 +++++
 file.h               |   1 +
 filesetup.c          |   2 +-
 fio.1                |   8 +
 fio.h                |   1 +
 goptions.c           |   8 +-
 init.c               |  49 ++++++
 libfio.c             |   2 +-
 options.c            |   9 ++
 options.h            |   2 +
 parse.h              |   5 +-
 server.h             |   2 +-
 thread_options.h     |   2 +
 17 files changed, 476 insertions(+), 147 deletions(-)
 create mode 100644 examples/waitfor.fio

---

Diff of recent changes:

diff --git a/HOWTO b/HOWTO
index b21d27e..7c24c1b 100644
--- a/HOWTO
+++ b/HOWTO
@@ -305,6 +305,16 @@ name=str	ASCII name of the job. This may be used to override the
 		special purpose of also signaling the start of a new
 		job.
 
+wait_for=str	Specifies the name of the already defined job to wait
+		for. Single waitee name only may be specified. If set, the job
+		won't be started until all workers of the waitee job are done.
+
+		Wait_for operates on the job name basis, so there are a few
+		limitations. First, the waitee must be defined prior to the
+		waiter job (meaning no forward references). Second, if a job
+		is being referenced as a waitee, it must have a unique name
+		(no duplicate waitees).
+
 description=str	Text description of the job. Doesn't do anything except
 		dump this text description when this job is run. It's
 		not parsed.
@@ -767,20 +777,14 @@ ioengine=str	Defines how the job issues io to the file. The following
 				defines engine specific options.
 
 			libhdfs	Read and write through Hadoop (HDFS).
-				The 'filename' option is used to specify host,
-				port of the hdfs name-node to connect. This
-				engine interprets offsets a little
+				This engine interprets offsets a little
 				differently. In HDFS, files once created
 				cannot be modified. So random writes are not
 				possible. To imitate this, libhdfs engine
-				expects bunch of small files to be created
-				over HDFS, and engine will randomly pick a
-				file out of those files based on the offset
-				generated by fio backend. (see the example
-				job file to create such files, use rw=write
-				option). Please note, you might want to set
-				necessary environment variables to work with
-				hdfs/libhdfs properly.
+				creates bunch of small files, and engine will
+				pick a file out of those files based on the 
+				offset enerated by fio backend. Each jobs uses
+				it's own connection to HDFS.
 
 			mtd	Read, write and erase an MTD character device
 				(e.g., /dev/mtd0). Discards are treated as
@@ -1785,11 +1789,13 @@ that defines them is selected.
 		If the job is a TCP listener or UDP reader, the hostname is not
 		used and must be omitted unless it is a valid UDP multicast
 		address.
+[libhdfs] namenode=str The host name or IP address of a HDFS cluster namenode to contact.
 
 [netsplice] port=int
 [net] port=int	The TCP or UDP port to bind to or connect to. If this is used
 with numjobs to spawn multiple instances of the same job type, then this will
 be the starting port number since fio will use a range of ports.
+[libhdfs] port=int	the listening port of the HFDS cluster namenode.
 
 [netsplice] interface=str
 [net] interface=str  The IP address of the network interface used to send or
@@ -1848,6 +1854,9 @@ be the starting port number since fio will use a range of ports.
 
 [mtd] skip_bad=bool	Skip operations against known bad blocks.
 
+[libhdfs] hdfsdirectory	libhdfs will create chunk in this HDFS directory
+[libhdfs] chunck_size	the size of the chunck to use for each file.
+
 
 6.0 Interpreting the output
 ---------------------------
diff --git a/backend.c b/backend.c
index 9920e63..bd94078 100644
--- a/backend.c
+++ b/backend.c
@@ -1978,6 +1978,32 @@ mounted:
 	return true;
 }
 
+static bool waitee_running(struct thread_data *me)
+{
+	const char *waitee = me->o.wait_for;
+	const char *self = me->o.name;
+	struct thread_data *td;
+	int i;
+
+	if (!waitee)
+		return false;
+
+	for_each_td(td, i) {
+		if (!strcmp(td->o.name, self) || strcmp(td->o.name, waitee))
+			continue;
+
+		if (td->runstate < TD_EXITED) {
+			dprint(FD_PROCESS, "%s fenced by %s(%s)\n",
+					self, td->o.name,
+					runstate_to_name(td->runstate));
+			return true;
+		}
+	}
+
+	dprint(FD_PROCESS, "%s: %s completed, can run\n", self, waitee);
+	return false;
+}
+
 /*
  * Main function for kicking off and reaping jobs, as needed.
  */
@@ -2101,6 +2127,12 @@ reap:
 				break;
 			}
 
+			if (waitee_running(td)) {
+				dprint(FD_PROCESS, "%s: waiting for %s\n",
+						td->o.name, td->o.wait_for);
+				continue;
+			}
+
 			init_disk_util(td);
 
 			td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED);
diff --git a/cconv.c b/cconv.c
index a476aad..6d8d0b3 100644
--- a/cconv.c
+++ b/cconv.c
@@ -25,6 +25,7 @@ static void free_thread_options_to_cpu(struct thread_options *o)
 {
 	free(o->description);
 	free(o->name);
+	free(o->wait_for);
 	free(o->directory);
 	free(o->filename);
 	free(o->filename_format);
@@ -54,6 +55,7 @@ void convert_thread_options_to_cpu(struct thread_options *o,
 
 	string_to_cpu(&o->description, top->description);
 	string_to_cpu(&o->name, top->name);
+	string_to_cpu(&o->wait_for, top->wait_for);
 	string_to_cpu(&o->directory, top->directory);
 	string_to_cpu(&o->filename, top->filename);
 	string_to_cpu(&o->filename_format, top->filename_format);
@@ -276,6 +278,7 @@ void convert_thread_options_to_net(struct thread_options_pack *top,
 
 	string_to_net(top->description, o->description);
 	string_to_net(top->name, o->name);
+	string_to_net(top->wait_for, o->wait_for);
 	string_to_net(top->directory, o->directory);
 	string_to_net(top->filename, o->filename);
 	string_to_net(top->filename_format, o->filename_format);
diff --git a/engines/libhdfs.c b/engines/libhdfs.c
index 658cd6a..f690b00 100644
--- a/engines/libhdfs.c
+++ b/engines/libhdfs.c
@@ -9,69 +9,123 @@
  *
  * thus, random reads and writes can also be achieved with this logic.
  *
- * NOTE: please set environment variables FIO_HDFS_BS and FIO_HDFS_FCOUNT
- * to appropriate value to work this engine properly
- *
  */
 
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <sys/uio.h>
-#include <errno.h>
-#include <assert.h>
+#include <math.h>
+#include <hdfs.h>
 
 #include "../fio.h"
 
-#include "hdfs.h"
+
+#define CHUNCK_NAME_LENGTH_MAX 80
+#define CHUNCK_CREATION_BUFFER_SIZE 65536
 
 struct hdfsio_data {
-	char host[256];
-	int port;
 	hdfsFS fs;
 	hdfsFile fp;
-	unsigned long fsbs;
-	unsigned long fscount;
-	unsigned long curr_file_id;
-	unsigned int numjobs;
-	unsigned int fid_correction;
+	uint64_t curr_file_id;
 };
 
-static int fio_hdfsio_setup_fs_params(struct hdfsio_data *hd)
-{
-	/* make sure that hdfsConnect is invoked before executing this function */
-	hdfsSetWorkingDirectory(hd->fs, "/.perftest");
-	hd->fp = hdfsOpenFile(hd->fs, ".fcount", O_RDONLY, 0, 0, 0);
-	if (hd->fp) {
-		hdfsRead(hd->fs, hd->fp, &(hd->fscount), sizeof(hd->fscount));
-		hdfsCloseFile(hd->fs, hd->fp);
-	}
-	hd->fp = hdfsOpenFile(hd->fs, ".fbs", O_RDONLY, 0, 0, 0);
-	if (hd->fp) {
-		hdfsRead(hd->fs, hd->fp, &(hd->fsbs), sizeof(hd->fsbs));
-		hdfsCloseFile(hd->fs, hd->fp);
-	}
+struct hdfsio_options {
+	void *pad;			/* needed because offset can't be 0 for a option defined used offsetof */
+	char *host;
+	char *directory;
+	unsigned int port;
+	unsigned int chunck_size;
+	unsigned int single_instance;
+	unsigned int use_direct;
+};
 
-	return 0;
+static struct fio_option options[] = {
+	{
+		.name	= "namenode",
+		.lname	= "hfds namenode",
+		.type	= FIO_OPT_STR_STORE,
+		.off1   = offsetof(struct hdfsio_options, host),
+		.def    = "localhost",
+		.help	= "Namenode of the HDFS cluster",
+		.category = FIO_OPT_C_ENGINE,
+		.group	= FIO_OPT_G_HDFS,
+	},
+	{
+		.name	= "hostname",
+		.lname	= "hfds namenode",
+		.type	= FIO_OPT_STR_STORE,
+		.off1   = offsetof(struct hdfsio_options, host),
+		.def    = "localhost",
+		.help	= "Namenode of the HDFS cluster",
+		.category = FIO_OPT_C_ENGINE,
+		.group	= FIO_OPT_G_HDFS,
+	},
+	{
+		.name	= "port",
+		.lname	= "hdfs namenode port",
+		.type	= FIO_OPT_INT,
+		.off1	= offsetof(struct hdfsio_options, port),
+		.def    = "9000",
+		.minval	= 1,
+		.maxval	= 65535,
+		.help	= "Port used by the HDFS cluster namenode",
+		.category = FIO_OPT_C_ENGINE,
+		.group	= FIO_OPT_G_HDFS,
+	},
+	{
+		.name	= "hdfsdirectory",
+		.lname	= "hfds directory",
+		.type	= FIO_OPT_STR_STORE,
+		.off1   = offsetof(struct hdfsio_options, directory),
+		.def    = "/",
+		.help	= "The HDFS directory where fio will create chuncks",
+		.category = FIO_OPT_C_ENGINE,
+		.group	= FIO_OPT_G_HDFS,
+	},
+	{
+		.name	= "chunck_size",
+		.type	= FIO_OPT_INT,
+		.off1	= offsetof(struct hdfsio_options, chunck_size),
+		.def    = "1048576",
+		.help	= "Size of individual chunck",
+		.category = FIO_OPT_C_ENGINE,
+		.group	= FIO_OPT_G_HDFS,
+	},
+	{
+		.name	= "single_instance",
+		.type	= FIO_OPT_BOOL,
+		.off1	= offsetof(struct hdfsio_options, single_instance),
+		.def    = "1",
+		.help	= "Use a single instance",
+		.category = FIO_OPT_C_ENGINE,
+		.group	= FIO_OPT_G_HDFS,
+	},
+	{
+		.name	= "hdfs_use_direct",
+		.type	= FIO_OPT_BOOL,
+		.off1	= offsetof(struct hdfsio_options, use_direct),
+		.def    = "0",
+		.help	= "Use readDirect instead of hdfsRead",
+		.category = FIO_OPT_C_ENGINE,
+		.group	= FIO_OPT_G_HDFS,
+	},
+	{
+		.name	= NULL,
+	},
+};
+
+
+static int get_chunck_name(char *dest, char *file_name, uint64_t chunk_id) {
+	return snprintf(dest, CHUNCK_NAME_LENGTH_MAX, "%s_%lu", file_name, chunk_id);
 }
 
 static int fio_hdfsio_prep(struct thread_data *td, struct io_u *io_u)
 {
-	struct hdfsio_data *hd;
-	hdfsFileInfo *fi;
+	struct hdfsio_options *options = td->eo;
+	struct hdfsio_data *hd = td->io_ops->data;
 	unsigned long f_id;
-	char fname[80];
-	int open_flags = 0;
-
-	hd = td->io_ops->data;
-
-	if (hd->curr_file_id == -1) {
-		/* see comment in fio_hdfsio_setup() function */
-		fio_hdfsio_setup_fs_params(hd);
-	}
+	char fname[CHUNCK_NAME_LENGTH_MAX];
+	int open_flags;
 
 	/* find out file id based on the offset generated by fio */
-	f_id = (io_u->offset / hd->fsbs) + hd->fid_correction;
+	f_id = floor(io_u->offset / options-> chunck_size);
 
 	if (f_id == hd->curr_file_id) {
 		/* file is already open */
@@ -79,46 +133,76 @@ static int fio_hdfsio_prep(struct thread_data *td, struct io_u *io_u)
 	}
 
 	if (hd->curr_file_id != -1) {
-		hdfsCloseFile(hd->fs, hd->fp);
+		if ( hdfsCloseFile(hd->fs, hd->fp) == -1) {
+			log_err("hdfs: unable to close file: %s\n", strerror(errno));
+			return errno;
+		}
+		hd->curr_file_id = -1;
 	}
 
-	if (io_u->ddir == DDIR_READ) {
+	if (io_u->ddir == DDIR_READ || io_u->ddir == DDIR_SYNC) {
 		open_flags = O_RDONLY;
 	} else if (io_u->ddir == DDIR_WRITE) {
 		open_flags = O_WRONLY;
 	} else {
 		log_err("hdfs: Invalid I/O Operation\n");
+		return 0;
+	}
+	
+	get_chunck_name(fname, io_u->file->file_name, f_id);
+	hd->fp = hdfsOpenFile(hd->fs, fname, open_flags, 0, 0,
+			      options->chunck_size);
+	if(hd->fp == NULL) {
+		log_err("hdfs: unable to open file: %s: %d\n", fname, strerror(errno));
+		return errno;
 	}
-
 	hd->curr_file_id = f_id;
-	do {
-		sprintf(fname, ".f%lu", f_id);
-		fi = hdfsGetPathInfo(hd->fs, fname);
-		if (fi->mSize >= hd->fsbs || io_u->ddir == DDIR_WRITE) {
-			/* file has enough data to read OR file is opened in write mode */
-			hd->fp =
-			    hdfsOpenFile(hd->fs, fname, open_flags, 0, 0,
-					 hd->fsbs);
-			if (hd->fp) {
-				break;
-			}
-		}
-		/* file is empty, so try next file for reading */
-		f_id = (f_id + 1) % hd->fscount;
-	} while (1);
 
 	return 0;
 }
 
-static int fio_io_end(struct thread_data *td, struct io_u *io_u, int ret)
+static int fio_hdfsio_queue(struct thread_data *td, struct io_u *io_u)
 {
+	struct hdfsio_data *hd = td->io_ops->data;
+	struct hdfsio_options *options = td->eo;
+	int ret;
+	unsigned long offset;
+	
+	offset = io_u->offset % options->chunck_size;
+	
+	if( (io_u->ddir == DDIR_READ || io_u->ddir == DDIR_WRITE) && 
+	     hdfsTell(hd->fs, hd->fp) != offset && hdfsSeek(hd->fs, hd->fp, offset) != 0 ) {
+		log_err("hdfs: seek failed: %s, are you doing random write smaller than chunck size ?\n", strerror(errno));
+		io_u->error = errno;
+		return FIO_Q_COMPLETED;
+	};
+
+	// do the IO
+	if (io_u->ddir == DDIR_READ) {
+		if (options->use_direct) {
+			ret = readDirect(hd->fs, hd->fp, io_u->xfer_buf, io_u->xfer_buflen);
+		} else {
+			ret = hdfsRead(hd->fs, hd->fp, io_u->xfer_buf, io_u->xfer_buflen);
+		}
+	} else if (io_u->ddir == DDIR_WRITE) {
+		ret = hdfsWrite(hd->fs, hd->fp, io_u->xfer_buf,
+				io_u->xfer_buflen);
+	} else if (io_u->ddir == DDIR_SYNC) {
+		ret = hdfsFlush(hd->fs, hd->fp);
+	} else {
+		log_err("hdfs: Invalid I/O Operation: %d\n", io_u->ddir);
+		ret = EINVAL;
+	}
+
+	// Check if the IO went fine, or is incomplete
 	if (ret != (int)io_u->xfer_buflen) {
 		if (ret >= 0) {
 			io_u->resid = io_u->xfer_buflen - ret;
 			io_u->error = 0;
 			return FIO_Q_COMPLETED;
-		} else
+		} else {
 			io_u->error = errno;
+		}
 	}
 
 	if (io_u->error)
@@ -127,107 +211,200 @@ static int fio_io_end(struct thread_data *td, struct io_u *io_u, int ret)
 	return FIO_Q_COMPLETED;
 }
 
-static int fio_hdfsio_queue(struct thread_data *td, struct io_u *io_u)
+int fio_hdfsio_open_file(struct thread_data *td, struct fio_file *f)
 {
-	struct hdfsio_data *hd;
-	int ret = 0;
-
-	hd = td->io_ops->data;
-
-	if (io_u->ddir == DDIR_READ) {
-		ret =
-		    hdfsRead(hd->fs, hd->fp, io_u->xfer_buf, io_u->xfer_buflen);
-	} else if (io_u->ddir == DDIR_WRITE) {
-		ret =
-		    hdfsWrite(hd->fs, hd->fp, io_u->xfer_buf,
-			      io_u->xfer_buflen);
-	} else {
-		log_err("hdfs: Invalid I/O Operation\n");
+	if (td->o.odirect) {
+		td->error = EINVAL;
+		return 0;
 	}
 
-	return fio_io_end(td, io_u, ret);
+	return 0;
 }
 
-int fio_hdfsio_open_file(struct thread_data *td, struct fio_file *f)
+int fio_hdfsio_close_file(struct thread_data *td, struct fio_file *f)
 {
-	struct hdfsio_data *hd;
-
-	hd = td->io_ops->data;
-	hd->fs = hdfsConnect(hd->host, hd->port);
-	hdfsSetWorkingDirectory(hd->fs, "/.perftest");
-	hd->fid_correction = (getpid() % hd->numjobs);
+	struct hdfsio_data *hd = td->io_ops->data;
 
+	if (hd->curr_file_id != -1) {
+		if ( hdfsCloseFile(hd->fs, hd->fp) == -1) {
+			log_err("hdfs: unable to close file: %s\n", strerror(errno));
+			return errno;
+		}
+		hd->curr_file_id = -1;
+	}
 	return 0;
 }
 
-int fio_hdfsio_close_file(struct thread_data *td, struct fio_file *f)
+static int fio_hdfsio_init(struct thread_data *td)
 {
-	struct hdfsio_data *hd;
-
-	hd = td->io_ops->data;
-	hdfsDisconnect(hd->fs);
+	struct hdfsio_options *options = td->eo;
+	struct hdfsio_data *hd = td->io_ops->data;
+	struct fio_file *f;
+	uint64_t j,k;
+	int i, failure = 0;
+	uint8_t buffer[CHUNCK_CREATION_BUFFER_SIZE];
+	uint64_t bytes_left;	
+	char fname[CHUNCK_NAME_LENGTH_MAX];	
+	hdfsFile fp;
+	hdfsFileInfo *fi;
+	tOffset fi_size;
+
+	for_each_file(td, f, i) {
+		k = 0;
+		for(j=0; j < f->real_file_size; j += options->chunck_size) {
+			get_chunck_name(fname, f->file_name, k++);
+			fi = hdfsGetPathInfo(hd->fs, fname);
+			fi_size = fi ? fi->mSize : 0;
+			// fill exist and is big enough, nothing to do
+			if( fi && fi_size >= options->chunck_size) {
+				continue;
+			}
+			fp = hdfsOpenFile(hd->fs, fname, O_WRONLY, 0, 0,
+					  options->chunck_size);
+			if(fp == NULL) {
+				failure = errno;
+				log_err("hdfs: unable to prepare file chunk %s: %s\n", fname, strerror(errno));
+				break;
+			}
+			bytes_left = options->chunck_size;
+			memset(buffer, 0, CHUNCK_CREATION_BUFFER_SIZE);
+			while( bytes_left > CHUNCK_CREATION_BUFFER_SIZE) {
+				if( hdfsWrite(hd->fs, fp, buffer, CHUNCK_CREATION_BUFFER_SIZE)
+				    != CHUNCK_CREATION_BUFFER_SIZE) {
+    					failure = errno;
+	    				log_err("hdfs: unable to prepare file chunk %s: %s\n", fname, strerror(errno));
+					break;
+				};
+				bytes_left -= CHUNCK_CREATION_BUFFER_SIZE;
+			}
+			if(bytes_left > 0) {
+				if( hdfsWrite(hd->fs, fp, buffer, bytes_left)
+				    != bytes_left) {
+					failure = errno;
+					break;
+				};
+			}
+			if( hdfsCloseFile(hd->fs, fp) != 0) {
+				failure = errno;
+				log_err("hdfs: unable to prepare file chunk %s: %s\n", fname, strerror(errno));
+				break;
+			}
+		}
+		if(failure) {
+			break;
+		}
+	}
+	
+	if( !failure ) {
+		fio_file_set_size_known(f);
+	}
 
-	return 0;
+	return failure;
 }
 
 static int fio_hdfsio_setup(struct thread_data *td)
 {
 	struct hdfsio_data *hd;
 	struct fio_file *f;
-	static unsigned int numjobs = 1;	/* atleast one job has to be there! */
-	numjobs = (td->o.numjobs > numjobs) ? td->o.numjobs : numjobs;
+	int i;
+	uint64_t file_size, total_file_size;
 
 	if (!td->io_ops->data) {
-		hd = malloc(sizeof(*hd));;
-
+		hd = malloc(sizeof(*hd));
 		memset(hd, 0, sizeof(*hd));
-		td->io_ops->data = hd;
+		
+		hd->curr_file_id = -1;
 
-		/* separate host and port from filename */
-		*(strchr(td->o.filename, ',')) = ' ';
-		sscanf(td->o.filename, "%s%d", hd->host, &(hd->port));
-
-		/* read fbs and fcount and based on that set f->real_file_size */
-		f = td->files[0];
-#if 0
-		/* IMHO, this should be done here instead of fio_hdfsio_prep()
-		 * but somehow calling it here doesn't seem to work,
-		 * some problem with libhdfs that needs to be debugged */
-		hd->fs = hdfsConnect(hd->host, hd->port);
-		fio_hdfsio_setup_fs_params(hd);
-		hdfsDisconnect(hd->fs);
-#else
-		/* so, as an alternate, using environment variables */
-		if (getenv("FIO_HDFS_FCOUNT") && getenv("FIO_HDFS_BS")) {
-			hd->fscount = atol(getenv("FIO_HDFS_FCOUNT"));
-			hd->fsbs = atol(getenv("FIO_HDFS_BS"));
-		} else {
-			log_err("FIO_HDFS_FCOUNT and/or FIO_HDFS_BS not set.\n");
-			return 1;
+		td->io_ops->data = hd;
+	}
+	
+	total_file_size = 0;
+	file_size = 0;
+
+	for_each_file(td, f, i) {
+		if(!td->o.file_size_low) {
+			file_size = floor(td->o.size / td->o.nr_files);
+			total_file_size += file_size;
 		}
-#endif
-		f->real_file_size = hd->fscount * hd->fsbs;
-
-		td->o.nr_files = 1;
-		hd->curr_file_id = -1;
-		hd->numjobs = numjobs;
-		fio_file_set_size_known(f);
+		else if (td->o.file_size_low == td->o.file_size_high)
+			file_size = td->o.file_size_low;
+		else {
+			file_size = get_rand_file_size(td);
+		}
+		f->real_file_size = file_size;
 	}
+	/* If the size doesn't divide nicely with the chunck size,
+	 * make the last files bigger.
+	 * Used only if filesize was not explicitely given
+	 */
+	if (!td->o.file_size_low && total_file_size < td->o.size) {
+		f->real_file_size += (td->o.size - total_file_size);
+	}
+
+	return 0;
+}
 
+static int fio_hdfsio_io_u_init(struct thread_data *td, struct io_u *io_u)
+{
+	struct hdfsio_data *hd = td->io_ops->data;
+	struct hdfsio_options *options = td->eo;
+	int failure;
+	struct hdfsBuilder *bld;
+
+	if (options->host == NULL || options->port == 0) {
+		log_err("hdfs: server not defined\n");
+		return EINVAL;
+	}
+	
+	bld = hdfsNewBuilder();
+	if (!bld) {
+		failure = errno;
+		log_err("hdfs: unable to allocate connect builder\n");
+		return failure;
+	}
+	hdfsBuilderSetNameNode(bld, options->host);
+	hdfsBuilderSetNameNodePort(bld, options->port);
+	if(! options->single_instance) {
+		hdfsBuilderSetForceNewInstance(bld);
+	}
+	hd->fs = hdfsBuilderConnect(bld);
+	
+	/* hdfsSetWorkingDirectory succeed on non existend directory */
+	if (hdfsExists(hd->fs, options->directory) < 0 || hdfsSetWorkingDirectory(hd->fs, options->directory) < 0) {
+		failure = errno;
+		log_err("hdfs: invalid working directory %s: %s\n", options->directory, strerror(errno));
+		return failure;
+	}
+	
 	return 0;
 }
 
+static void fio_hdfsio_io_u_free(struct thread_data *td, struct io_u *io_u)
+{
+	struct hdfsio_data *hd = td->io_ops->data;
+
+	if (hd->fs && hdfsDisconnect(hd->fs) < 0) {
+		log_err("hdfs: disconnect failed: %d\n", errno);
+	}
+}
+
 static struct ioengine_ops ioengine_hdfs = {
 	.name = "libhdfs",
 	.version = FIO_IOOPS_VERSION,
+	.flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_NODISKUTIL,
 	.setup = fio_hdfsio_setup,
+	.init = fio_hdfsio_init,
 	.prep = fio_hdfsio_prep,
 	.queue = fio_hdfsio_queue,
 	.open_file = fio_hdfsio_open_file,
 	.close_file = fio_hdfsio_close_file,
-	.flags = FIO_SYNCIO,
+	.io_u_init = fio_hdfsio_io_u_init,
+	.io_u_free = fio_hdfsio_io_u_free,
+	.option_struct_size	= sizeof(struct hdfsio_options),
+	.options		= options,
 };
 
+
 static void fio_init fio_hdfsio_register(void)
 {
 	register_ioengine(&ioengine_hdfs);
diff --git a/examples/waitfor.fio b/examples/waitfor.fio
new file mode 100644
index 0000000..95fad00
--- /dev/null
+++ b/examples/waitfor.fio
@@ -0,0 +1,35 @@
+[global]
+threads=1
+group_reporting=1
+filename=/tmp/data
+filesize=128m
+
+[writers]
+rw=write
+bs=128k
+numjobs=4
+runtime=10
+
+[readers]
+new_group
+wait_for=writers
+rw=randread
+bs=4k
+numjobs=4
+runtime=10
+
+[writers2]
+new_group
+wait_for=readers
+rw=randwrite
+bs=4k
+numjobs=4
+runtime=10
+
+[readers2]
+new_group
+wait_for=writers2
+rw=randread
+bs=4k
+numjobs=4
+runtime=10
diff --git a/file.h b/file.h
index d5595c1..a631766 100644
--- a/file.h
+++ b/file.h
@@ -180,6 +180,7 @@ extern int __must_check generic_close_file(struct thread_data *, struct fio_file
 extern int __must_check generic_get_file_size(struct thread_data *, struct fio_file *);
 extern int __must_check file_lookup_open(struct fio_file *f, int flags);
 extern int __must_check pre_read_files(struct thread_data *);
+extern unsigned long long get_rand_file_size(struct thread_data *td);
 extern int add_file(struct thread_data *, const char *, int, int);
 extern int add_file_exclusive(struct thread_data *, const char *);
 extern void get_file(struct fio_file *);
diff --git a/filesetup.c b/filesetup.c
index 7666754..a821632 100644
--- a/filesetup.c
+++ b/filesetup.c
@@ -264,7 +264,7 @@ error:
 	return ret;
 }
 
-static unsigned long long get_rand_file_size(struct thread_data *td)
+unsigned long long get_rand_file_size(struct thread_data *td)
 {
 	unsigned long long ret, sized;
 	uint64_t frand_max;
diff --git a/fio.1 b/fio.1
index 4fe1be2..7bdfea3 100644
--- a/fio.1
+++ b/fio.1
@@ -180,6 +180,14 @@ a ':' character.
 May be used to override the job name.  On the command line, this parameter
 has the special purpose of signalling the start of a new job.
 .TP
+.BI wait_for \fR=\fPstr
+Specifies the name of the already defined job to wait for. Single waitee name
+only may be specified. If set, the job won't be started until all workers of
+the waitee job are done.  Wait_for operates on the job name basis, so there are
+a few limitations. First, the waitee must be defined prior to the waiter job
+(meaning no forward references). Second, if a job is being referenced as a
+waitee, it must have a unique name (no duplicate waitees).
+.TP
 .BI description \fR=\fPstr
 Human-readable description of the job. It is printed when the job is run, but
 otherwise has no special purpose.
diff --git a/fio.h b/fio.h
index ddc29db..66211e9 100644
--- a/fio.h
+++ b/fio.h
@@ -542,6 +542,7 @@ enum {
 extern void td_set_runstate(struct thread_data *, int);
 extern int td_bump_runstate(struct thread_data *, int);
 extern void td_restore_runstate(struct thread_data *, int);
+extern const char *runstate_to_name(int runstate);
 
 /*
  * Allow 60 seconds for a job to quit on its own, otherwise reap with
diff --git a/goptions.c b/goptions.c
index c01b6cc..20a17d1 100644
--- a/goptions.c
+++ b/goptions.c
@@ -92,9 +92,9 @@ struct gopt_job_view {
 static GNode *gopt_dep_tree;
 
 static GtkWidget *gopt_get_group_frame(struct gopt_job_view *gjv,
-				       GtkWidget *box, unsigned int groupmask)
+				       GtkWidget *box, uint64_t groupmask)
 {
-	unsigned int mask, group;
+	uint64_t mask, group;
 	struct opt_group *og;
 	GtkWidget *frame, *hbox;
 	struct gopt_frame_widget *gfw;
@@ -1135,7 +1135,7 @@ static void gopt_add_options(struct gopt_job_view *gjv,
 	 */
 	for (i = 0; fio_options[i].name; i++) {
 		struct fio_option *o = &fio_options[i];
-		unsigned int mask = o->category;
+		uint64_t mask = o->category;
 		struct opt_group *og;
 
 		while ((og = opt_group_from_mask(&mask)) != NULL) {
@@ -1189,7 +1189,7 @@ static void gopt_add_group_tabs(GtkWidget *notebook, struct gopt_job_view *gjv)
 
 	i = 0;
 	do {
-		unsigned int mask = (1U << i);
+		uint64_t mask = (1ULL << i);
 
 		og = opt_group_from_mask(&mask);
 		if (!og)
diff --git a/init.c b/init.c
index 8773138..991fa1c 100644
--- a/init.c
+++ b/init.c
@@ -1217,6 +1217,49 @@ static void gen_log_name(char *name, size_t size, const char *logtype,
 		snprintf(name, size, "%s_%s.%s", logname, logtype, suf);
 }
 
+static int check_waitees(char *waitee)
+{
+	struct thread_data *td;
+	int i, ret = 0;
+
+	for_each_td(td, i) {
+		if (td->subjob_number)
+			continue;
+
+		ret += !strcmp(td->o.name, waitee);
+	}
+
+	return ret;
+}
+
+static bool wait_for_ok(const char *jobname, struct thread_options *o)
+{
+	int nw;
+
+	if (!o->wait_for)
+		return true;
+
+	if (!strcmp(jobname, o->wait_for)) {
+		log_err("%s: a job cannot wait for itself (wait_for=%s).\n",
+				jobname, o->wait_for);
+		return false;
+	}
+
+	if (!(nw = check_waitees(o->wait_for))) {
+		log_err("%s: waitee job %s unknown.\n", jobname, o->wait_for);
+		return false;
+	}
+
+	if (nw > 1) {
+		log_err("%s: multiple waitees %s found,\n"
+			"please avoid duplicates when using wait_for option.\n",
+				jobname, o->wait_for);
+		return false;
+	}
+
+	return true;
+}
+
 /*
  * Adds a job to the list of things todo. Sanitizes the various options
  * to make sure we don't have conflicts, and initializes various
@@ -1273,6 +1316,12 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num,
 	if (fixup_options(td))
 		goto err;
 
+	/*
+	 * Belongs to fixup_options, but o->name is not necessarily set as yet
+	 */
+	if (!wait_for_ok(jobname, o))
+		goto err;
+
 	flow_init_job(td);
 
 	/*
diff --git a/libfio.c b/libfio.c
index 6c74852..c626d15 100644
--- a/libfio.c
+++ b/libfio.c
@@ -190,7 +190,7 @@ static const char *td_runstates[] = {
 	"REAPED",
 };
 
-static const char *runstate_to_name(int runstate)
+const char *runstate_to_name(int runstate)
 {
 	compiletime_assert(TD_LAST == 12, "td runstate list");
 	if (runstate >= 0 && runstate < TD_LAST)
diff --git a/options.c b/options.c
index 8494713..49d6600 100644
--- a/options.c
+++ b/options.c
@@ -1229,6 +1229,15 @@ struct fio_option fio_options[FIO_MAX_OPTS] = {
 		.group	= FIO_OPT_G_DESC,
 	},
 	{
+		.name	= "wait_for",
+		.lname	= "Waitee name",
+		.type	= FIO_OPT_STR_STORE,
+		.off1	= td_var_offset(wait_for),
+		.help	= "Name of the job this one wants to wait for before starting",
+		.category = FIO_OPT_C_GENERAL,
+		.group	= FIO_OPT_G_DESC,
+	},
+	{
 		.name	= "filename",
 		.lname	= "Filename(s)",
 		.type	= FIO_OPT_STR_STORE,
diff --git a/options.h b/options.h
index d37b162..13b534a 100644
--- a/options.h
+++ b/options.h
@@ -116,6 +116,7 @@ enum opt_category_group {
         __FIO_OPT_G_RBD,
         __FIO_OPT_G_GFAPI,
         __FIO_OPT_G_MTD,
+	__FIO_OPT_G_HDFS,
 	__FIO_OPT_G_NR,
 
 	FIO_OPT_G_RATE		= (1ULL << __FIO_OPT_G_RATE),
@@ -149,6 +150,7 @@ enum opt_category_group {
 	FIO_OPT_G_RBD		= (1ULL << __FIO_OPT_G_RBD),
 	FIO_OPT_G_GFAPI		= (1ULL << __FIO_OPT_G_GFAPI),
 	FIO_OPT_G_MTD		= (1ULL << __FIO_OPT_G_MTD),
+	FIO_OPT_G_HDFS		= (1ULL << __FIO_OPT_G_HDFS),
 	FIO_OPT_G_INVALID	= (1ULL << __FIO_OPT_G_NR),
 };
 
diff --git a/parse.h b/parse.h
index 3ba8047..77450ef 100644
--- a/parse.h
+++ b/parse.h
@@ -1,6 +1,7 @@
 #ifndef FIO_PARSE_H
 #define FIO_PARSE_H
 
+#include <inttypes.h>
 #include "flist.h"
 
 /*
@@ -69,8 +70,8 @@ struct fio_option {
 	int (*verify)(struct fio_option *, void *);
 	const char *prof_name;		/* only valid for specific profile */
 	void *prof_opts;
-	unsigned int category;		/* what type of option */
-	unsigned int group;		/* who to group with */
+	uint64_t category;		/* what type of option */
+	uint64_t group;			/* who to group with */
 	void *gui_data;
 	int is_seconds;			/* time value with seconds base */
 	int is_time;			/* time based value */
diff --git a/server.h b/server.h
index 5a59d07..9205ae6 100644
--- a/server.h
+++ b/server.h
@@ -38,7 +38,7 @@ struct fio_net_cmd_reply {
 };
 
 enum {
-	FIO_SERVER_VER			= 50,
+	FIO_SERVER_VER			= 51,
 
 	FIO_SERVER_MAX_FRAGMENT_PDU	= 1024,
 	FIO_SERVER_MAX_CMD_MB		= 2048,
diff --git a/thread_options.h b/thread_options.h
index 6ae0335..858f307 100644
--- a/thread_options.h
+++ b/thread_options.h
@@ -40,6 +40,7 @@ struct thread_options {
 	uint64_t set_options[NR_OPTS_SZ];
 	char *description;
 	char *name;
+	char *wait_for;
 	char *directory;
 	char *filename;
 	char *filename_format;
@@ -289,6 +290,7 @@ struct thread_options_pack {
 	uint64_t set_options[NR_OPTS_SZ];
 	uint8_t description[FIO_TOP_STR_MAX];
 	uint8_t name[FIO_TOP_STR_MAX];
+	uint8_t wait_for[FIO_TOP_STR_MAX];
 	uint8_t directory[FIO_TOP_STR_MAX];
 	uint8_t filename[FIO_TOP_STR_MAX];
 	uint8_t filename_format[FIO_TOP_STR_MAX];
--
To unsubscribe from this list: send the line "unsubscribe fio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html



[Index of Archives]     [Linux Kernel]     [Linux SCSI]     [Linux IDE]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux SCSI]

  Powered by Linux