view contrib/chg/hgclient.c @ 30818:4c0a5a256ae8

localrepo: experimental support for non-zlib revlog compression

The final part of integrating the compression manager APIs into
revlog storage is the plumbing for repositories to advertise they
are using non-zlib storage and for revlogs to instantiate a non-zlib
compression engine.

The main intent of the compression manager work was to zstd all
of the things. Adding zstd to revlogs has proved to be more involved
than other places because revlogs are... special. Very small inputs
and the use of delta chains (which are themselves a form of
compression) are a completely different use case from streaming
compression, which bundles and the wire protocol employ. I've
conducted numerous experiments with zstd in revlogs and have yet
to formalize compression settings and a storage architecture that
I'm confident I won't regret later. In other words, I'm not yet
ready to commit to a new mechanism for using zstd - or any other
compression format - in revlogs.

That being said, having some support for zstd (and other compression
formats) in revlogs in core is beneficial. It can allow others to
conduct experiments.

This patch introduces *highly experimental* support for non-zlib
compression formats in revlogs. Introduced is a config option to
control which compression engine to use. Also introduced is a namespace
of "exp-compression-*" requirements to denote support for non-zlib
compression in revlogs. I've prefixed the namespace with "exp-"
(short for "experimental") because I'm not confident of the
requirements "schema" and in no way want to give the illusion of
supporting these requirements in the future. I fully intend to drop
support for these requirements once we figure out what we're doing
with zstd in revlogs.

A good portion of the patch is teaching the requirements system
about registered compression engines and passing the requested
compression engine as an opener option so revlogs can instantiate
the proper compression engine for new operations.

That's a verbose way of saying "we can now use zstd in revlogs!"

On an `hg pull` conversion of the mozilla-unified repo with no extra
redelta settings (like aggressivemergedeltas), we can see the impact
of zstd vs zlib in revlogs:

$ hg perfrevlogchunks -c
! chunk
! wall 2.032052 comb 2.040000 user 1.990000 sys 0.050000 (best of 5)
! wall 1.866360 comb 1.860000 user 1.820000 sys 0.040000 (best of 6)

! chunk batch
! wall 1.877261 comb 1.870000 user 1.860000 sys 0.010000 (best of 6)
! wall 1.705410 comb 1.710000 user 1.690000 sys 0.020000 (best of 6)

$ hg perfrevlogchunks -m
! chunk
! wall 2.721427 comb 2.720000 user 2.640000 sys 0.080000 (best of 4)
! wall 2.035076 comb 2.030000 user 1.950000 sys 0.080000 (best of 5)

! chunk batch
! wall 2.614561 comb 2.620000 user 2.580000 sys 0.040000 (best of 4)
! wall 1.910252 comb 1.910000 user 1.880000 sys 0.030000 (best of 6)

$ hg perfrevlog -c -d 1
! wall 4.812885 comb 4.820000 user 4.800000 sys 0.020000 (best of 3)
! wall 4.699621 comb 4.710000 user 4.700000 sys 0.010000 (best of 3)

$ hg perfrevlog -m -d 1000
! wall 34.252800 comb 34.250000 user 33.730000 sys 0.520000 (best of 3)
! wall 24.094999 comb 24.090000 user 23.320000 sys 0.770000 (best of 3)

Only modest wins for the changelog. But manifest reading is
significantly faster. What's going on?

One reason might be data volume. zstd decompresses faster. So given
more bytes, it will put more distance between it and zlib.

Another reason is size. In the current design, zstd revlogs are
*larger*:

debugcreatestreamclonebundle (size in bytes)
zlib: 1,638,852,492
zstd: 1,680,601,332

I haven't investigated this fully, but I reckon a significant cause of
larger revlogs is that the zstd frame/header has more bytes than
zlib's. For very small inputs or data that doesn't compress well, we'll
tend to store more uncompressed chunks than with zlib (because the
compressed size isn't smaller than original). This will make revlog
reading faster because it is doing less decompression.

Moving on to bundle performance:

$ hg bundle -a -t none-v2 (total CPU time)
zlib: 102.79s
zstd:  97.75s

So, marginal CPU decrease for reading all chunks in all revlogs
(this is somewhat disappointing).

$ hg bundle -a -t <engine>-v2 (total CPU time)
zlib: 191.59s
zstd: 115.36s

This last test effectively measures the difference between zlib->zlib
and zstd->zstd for revlogs to bundle. This is a rough approximation of
what a server does during `hg clone`.

There are some promising results for zstd. But not enough for me to
feel comfortable advertising it to users. We'll get there...

/*
 * A command server client that uses Unix domain socket
 *
 * Copyright (c) 2011 Yuya Nishihara <[e-mail]>
 *
 * This software may be used and distributed according to the terms of the
 * GNU General Public License version 2 or any later version.
 */

#include <arpa/inet.h>  /* for ntohl(), htonl() */
#include <assert.h>
#include <ctype.h>
#include <errno.h>
#include <fcntl.h>
#include <signal.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/un.h>
#include <unistd.h>

#include "hgclient.h"
#include "procutil.h"
#include "util.h"

enum {
	CAP_GETENCODING = 0x0001,
	CAP_RUNCOMMAND = 0x0002,
	/* cHg extension: */
	CAP_ATTACHIO = 0x0100,
	CAP_CHDIR = 0x0200,
	CAP_SETENV = 0x0800,
	CAP_SETUMASK = 0x1000,
	CAP_VALIDATE = 0x2000,
	CAP_SETPROCNAME = 0x4000,
};

typedef struct {
	const char *name;
	unsigned int flag;
} cappair_t;

static const cappair_t captable[] = {
	{"getencoding", CAP_GETENCODING},
	{"runcommand", CAP_RUNCOMMAND},
	{"attachio", CAP_ATTACHIO},
	{"chdir", CAP_CHDIR},
	{"setenv", CAP_SETENV},
	{"setumask", CAP_SETUMASK},
	{"validate", CAP_VALIDATE},
	{"setprocname", CAP_SETPROCNAME},
	{NULL, 0},  /* terminator */
};

typedef struct {
	char ch;
	char *data;
	size_t maxdatasize;
	size_t datasize;
} context_t;

struct hgclient_tag_ {
	int sockfd;
	pid_t pgid;
	pid_t pid;
	context_t ctx;
	unsigned int capflags;
};

static const size_t defaultdatasize = 4096;

static void attachio(hgclient_t *hgc);

static void initcontext(context_t *ctx)
{
	ctx->ch = '\0';
	ctx->data = malloc(defaultdatasize);
	ctx->maxdatasize = (ctx->data) ? defaultdatasize : 0;
	ctx->datasize = 0;
	debugmsg("initialize context buffer with size %zu", ctx->maxdatasize);
}

static void enlargecontext(context_t *ctx, size_t newsize)
{
	if (newsize <= ctx->maxdatasize)
		return;

	newsize = defaultdatasize
		* ((newsize + defaultdatasize - 1) / defaultdatasize);
	ctx->data = reallocx(ctx->data, newsize);
	ctx->maxdatasize = newsize;
	debugmsg("enlarge context buffer to %zu", ctx->maxdatasize);
}

static void freecontext(context_t *ctx)
{
	debugmsg("free context buffer");
	free(ctx->data);
	ctx->data = NULL;
	ctx->maxdatasize = 0;
	ctx->datasize = 0;
}

/* Read channeled response from cmdserver */
static void readchannel(hgclient_t *hgc)
{
	assert(hgc);

	ssize_t rsize = recv(hgc->sockfd, &hgc->ctx.ch, sizeof(hgc->ctx.ch), 0);
	if (rsize != sizeof(hgc->ctx.ch)) {
		/* server would have exception and traceback would be printed */
		debugmsg("failed to read channel");
		exit(255);
	}

	uint32_t datasize_n;
	rsize = recv(hgc->sockfd, &datasize_n, sizeof(datasize_n), 0);
	if (rsize != sizeof(datasize_n))
		abortmsg("failed to read data size");

	/* datasize denotes the maximum size to write if input request */
	hgc->ctx.datasize = ntohl(datasize_n);
	enlargecontext(&hgc->ctx, hgc->ctx.datasize);

	if (isupper(hgc->ctx.ch) && hgc->ctx.ch != 'S')
		return;  /* assumes input request */

	size_t cursize = 0;
	while (cursize < hgc->ctx.datasize) {
		rsize = recv(hgc->sockfd, hgc->ctx.data + cursize,
			     hgc->ctx.datasize - cursize, 0);
		if (rsize < 1)
			abortmsg("failed to read data block");
		cursize += rsize;
	}
}

static void sendall(int sockfd, const void *data, size_t datasize)
{
	const char *p = data;
	const char *const endp = p + datasize;
	while (p < endp) {
		ssize_t r = send(sockfd, p, endp - p, 0);
		if (r < 0)
			abortmsgerrno("cannot communicate");
		p += r;
	}
}

/* Write lengh-data block to cmdserver */
static void writeblock(const hgclient_t *hgc)
{
	assert(hgc);

	const uint32_t datasize_n = htonl(hgc->ctx.datasize);
	sendall(hgc->sockfd, &datasize_n, sizeof(datasize_n));

	sendall(hgc->sockfd, hgc->ctx.data, hgc->ctx.datasize);
}

static void writeblockrequest(const hgclient_t *hgc, const char *chcmd)
{
	debugmsg("request %s, block size %zu", chcmd, hgc->ctx.datasize);

	char buf[strlen(chcmd) + 1];
	memcpy(buf, chcmd, sizeof(buf) - 1);
	buf[sizeof(buf) - 1] = '\n';
	sendall(hgc->sockfd, buf, sizeof(buf));

	writeblock(hgc);
}

/* Build '\0'-separated list of args. argsize < 0 denotes that args are
 * terminated by NULL. */
static void packcmdargs(context_t *ctx, const char *const args[],
			ssize_t argsize)
{
	ctx->datasize = 0;
	const char *const *const end = (argsize >= 0) ? args + argsize : NULL;
	for (const char *const *it = args; it != end && *it; ++it) {
		const size_t n = strlen(*it) + 1;  /* include '\0' */
		enlargecontext(ctx, ctx->datasize + n);
		memcpy(ctx->data + ctx->datasize, *it, n);
		ctx->datasize += n;
	}

	if (ctx->datasize > 0)
		--ctx->datasize;  /* strip last '\0' */
}

/* Extract '\0'-separated list of args to new buffer, terminated by NULL */
static const char **unpackcmdargsnul(const context_t *ctx)
{
	const char **args = NULL;
	size_t nargs = 0, maxnargs = 0;
	const char *s = ctx->data;
	const char *e = ctx->data + ctx->datasize;
	for (;;) {
		if (nargs + 1 >= maxnargs) {  /* including last NULL */
			maxnargs += 256;
			args = reallocx(args, maxnargs * sizeof(args[0]));
		}
		args[nargs] = s;
		nargs++;
		s = memchr(s, '\0', e - s);
		if (!s)
			break;
		s++;
	}
	args[nargs] = NULL;
	return args;
}

static void handlereadrequest(hgclient_t *hgc)
{
	context_t *ctx = &hgc->ctx;
	size_t r = fread(ctx->data, sizeof(ctx->data[0]), ctx->datasize, stdin);
	ctx->datasize = r;
	writeblock(hgc);
}

/* Read single-line */
static void handlereadlinerequest(hgclient_t *hgc)
{
	context_t *ctx = &hgc->ctx;
	if (!fgets(ctx->data, ctx->datasize, stdin))
		ctx->data[0] = '\0';
	ctx->datasize = strlen(ctx->data);
	writeblock(hgc);
}

/* Execute the requested command and write exit code */
static void handlesystemrequest(hgclient_t *hgc)
{
	context_t *ctx = &hgc->ctx;
	enlargecontext(ctx, ctx->datasize + 1);
	ctx->data[ctx->datasize] = '\0';  /* terminate last string */

	const char **args = unpackcmdargsnul(ctx);
	if (!args[0] || !args[1] || !args[2])
		abortmsg("missing type or command or cwd in system request");
	if (strcmp(args[0], "system") == 0) {
		debugmsg("run '%s' at '%s'", args[1], args[2]);
		int32_t r = runshellcmd(args[1], args + 3, args[2]);
		free(args);

		uint32_t r_n = htonl(r);
		memcpy(ctx->data, &r_n, sizeof(r_n));
		ctx->datasize = sizeof(r_n);
		writeblock(hgc);
	} else if (strcmp(args[0], "pager") == 0) {
		setuppager(args[1]);
		if (hgc->capflags & CAP_ATTACHIO)
			attachio(hgc);
		/* unblock the server */
		static const char emptycmd[] = "\n";
		sendall(hgc->sockfd, emptycmd, sizeof(emptycmd) - 1);
	} else {
		abortmsg("unknown type in system request: %s", args[0]);
	}
}

/* Read response of command execution until receiving 'r'-esult */
static void handleresponse(hgclient_t *hgc)
{
	for (;;) {
		readchannel(hgc);
		context_t *ctx = &hgc->ctx;
		debugmsg("response read from channel %c, size %zu",
			 ctx->ch, ctx->datasize);
		switch (ctx->ch) {
		case 'o':
			fwrite(ctx->data, sizeof(ctx->data[0]), ctx->datasize,
			       stdout);
			break;
		case 'e':
			fwrite(ctx->data, sizeof(ctx->data[0]), ctx->datasize,
			       stderr);
			break;
		case 'd':
			/* assumes last char is '\n' */
			ctx->data[ctx->datasize - 1] = '\0';
			debugmsg("server: %s", ctx->data);
			break;
		case 'r':
			return;
		case 'I':
			handlereadrequest(hgc);
			break;
		case 'L':
			handlereadlinerequest(hgc);
			break;
		case 'S':
			handlesystemrequest(hgc);
			break;
		default:
			if (isupper(ctx->ch))
				abortmsg("cannot handle response (ch = %c)",
					 ctx->ch);
		}
	}
}

static unsigned int parsecapabilities(const char *s, const char *e)
{
	unsigned int flags = 0;
	while (s < e) {
		const char *t = strchr(s, ' ');
		if (!t || t > e)
			t = e;
		const cappair_t *cap;
		for (cap = captable; cap->flag; ++cap) {
			size_t n = t - s;
			if (strncmp(s, cap->name, n) == 0 &&
			    strlen(cap->name) == n) {
				flags |= cap->flag;
				break;
			}
		}
		s = t + 1;
	}
	return flags;
}

static void readhello(hgclient_t *hgc)
{
	readchannel(hgc);
	context_t *ctx = &hgc->ctx;
	if (ctx->ch != 'o') {
		char ch = ctx->ch;
		if (ch == 'e') {
			/* write early error and will exit */
			fwrite(ctx->data, sizeof(ctx->data[0]), ctx->datasize,
			       stderr);
			handleresponse(hgc);
		}
		abortmsg("unexpected channel of hello message (ch = %c)", ch);
	}
	enlargecontext(ctx, ctx->datasize + 1);
	ctx->data[ctx->datasize] = '\0';
	debugmsg("hello received: %s (size = %zu)", ctx->data, ctx->datasize);

	const char *s = ctx->data;
	const char *const dataend = ctx->data + ctx->datasize;
	while (s < dataend) {
		const char *t = strchr(s, ':');
		if (!t || t[1] != ' ')
			break;
		const char *u = strchr(t + 2, '\n');
		if (!u)
			u = dataend;
		if (strncmp(s, "capabilities:", t - s + 1) == 0) {
			hgc->capflags = parsecapabilities(t + 2, u);
		} else if (strncmp(s, "pgid:", t - s + 1) == 0) {
			hgc->pgid = strtol(t + 2, NULL, 10);
		} else if (strncmp(s, "pid:", t - s + 1) == 0) {
			hgc->pid = strtol(t + 2, NULL, 10);
		}
		s = u + 1;
	}
	debugmsg("capflags=0x%04x, pid=%d", hgc->capflags, hgc->pid);
}

static void updateprocname(hgclient_t *hgc)
{
	int r = snprintf(hgc->ctx.data, hgc->ctx.maxdatasize,
			"chg[worker/%d]", (int)getpid());
	if (r < 0 || (size_t)r >= hgc->ctx.maxdatasize)
		abortmsg("insufficient buffer to write procname (r = %d)", r);
	hgc->ctx.datasize = (size_t)r;
	writeblockrequest(hgc, "setprocname");
}

static void attachio(hgclient_t *hgc)
{
	debugmsg("request attachio");
	static const char chcmd[] = "attachio\n";
	sendall(hgc->sockfd, chcmd, sizeof(chcmd) - 1);
	readchannel(hgc);
	context_t *ctx = &hgc->ctx;
	if (ctx->ch != 'I')
		abortmsg("unexpected response for attachio (ch = %c)", ctx->ch);

	static const int fds[3] = {STDIN_FILENO, STDOUT_FILENO, STDERR_FILENO};
	struct msghdr msgh;
	memset(&msgh, 0, sizeof(msgh));
	struct iovec iov = {ctx->data, ctx->datasize};  /* dummy payload */
	msgh.msg_iov = &iov;
	msgh.msg_iovlen = 1;
	char fdbuf[CMSG_SPACE(sizeof(fds))];
	msgh.msg_control = fdbuf;
	msgh.msg_controllen = sizeof(fdbuf);
	struct cmsghdr *cmsg = CMSG_FIRSTHDR(&msgh);
	cmsg->cmsg_level = SOL_SOCKET;
	cmsg->cmsg_type = SCM_RIGHTS;
	cmsg->cmsg_len = CMSG_LEN(sizeof(fds));
	memcpy(CMSG_DATA(cmsg), fds, sizeof(fds));
	msgh.msg_controllen = cmsg->cmsg_len;
	ssize_t r = sendmsg(hgc->sockfd, &msgh, 0);
	if (r < 0)
		abortmsgerrno("sendmsg failed");

	handleresponse(hgc);
	int32_t n;
	if (ctx->datasize != sizeof(n))
		abortmsg("unexpected size of attachio result");
	memcpy(&n, ctx->data, sizeof(n));
	n = ntohl(n);
	if (n != sizeof(fds) / sizeof(fds[0]))
		abortmsg("failed to send fds (n = %d)", n);
}

static void chdirtocwd(hgclient_t *hgc)
{
	if (!getcwd(hgc->ctx.data, hgc->ctx.maxdatasize))
		abortmsgerrno("failed to getcwd");
	hgc->ctx.datasize = strlen(hgc->ctx.data);
	writeblockrequest(hgc, "chdir");
}

static void forwardumask(hgclient_t *hgc)
{
	mode_t mask = umask(0);
	umask(mask);

	static const char command[] = "setumask\n";
	sendall(hgc->sockfd, command, sizeof(command) - 1);
	uint32_t data = htonl(mask);
	sendall(hgc->sockfd, &data, sizeof(data));
}

/*!
 * Open connection to per-user cmdserver
 *
 * If no background server running, returns NULL.
 */
hgclient_t *hgc_open(const char *sockname)
{
	int fd = socket(AF_UNIX, SOCK_STREAM, 0);
	if (fd < 0)
		abortmsgerrno("cannot create socket");

	/* don't keep fd on fork(), so that it can be closed when the parent
	 * process get terminated. */
	fsetcloexec(fd);

	struct sockaddr_un addr;
	addr.sun_family = AF_UNIX;

	/* use chdir to workaround small sizeof(sun_path) */
	int bakfd = -1;
	const char *basename = sockname;
	{
		const char *split = strrchr(sockname, '/');
		if (split && split != sockname) {
			if (split[1] == '\0')
				abortmsg("sockname cannot end with a slash");
			size_t len = split - sockname;
			char sockdir[len + 1];
			memcpy(sockdir, sockname, len);
			sockdir[len] = '\0';

			bakfd = open(".", O_DIRECTORY);
			if (bakfd == -1)
				abortmsgerrno("cannot open cwd");

			int r = chdir(sockdir);
			if (r != 0)
				abortmsgerrno("cannot chdir %s", sockdir);

			basename = split + 1;
		}
	}
	if (strlen(basename) >= sizeof(addr.sun_path))
		abortmsg("sockname is too long: %s", basename);
	strncpy(addr.sun_path, basename, sizeof(addr.sun_path));
	addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';

	/* real connect */
	int r = connect(fd, (struct sockaddr *)&addr, sizeof(addr));
	if (r < 0) {
		if (errno != ENOENT && errno != ECONNREFUSED)
			abortmsgerrno("cannot connect to %s", sockname);
	}
	if (bakfd != -1) {
		fchdirx(bakfd);
		close(bakfd);
	}
	if (r < 0) {
		close(fd);
		return NULL;
	}
	debugmsg("connected to %s", addr.sun_path);

	hgclient_t *hgc = mallocx(sizeof(hgclient_t));
	memset(hgc, 0, sizeof(*hgc));
	hgc->sockfd = fd;
	initcontext(&hgc->ctx);

	readhello(hgc);
	if (!(hgc->capflags & CAP_RUNCOMMAND))
		abortmsg("insufficient capability: runcommand");
	if (hgc->capflags & CAP_SETPROCNAME)
		updateprocname(hgc);
	if (hgc->capflags & CAP_ATTACHIO)
		attachio(hgc);
	if (hgc->capflags & CAP_CHDIR)
		chdirtocwd(hgc);
	if (hgc->capflags & CAP_SETUMASK)
		forwardumask(hgc);

	return hgc;
}

/*!
 * Close connection and free allocated memory
 */
void hgc_close(hgclient_t *hgc)
{
	assert(hgc);
	freecontext(&hgc->ctx);
	close(hgc->sockfd);
	free(hgc);
}

pid_t hgc_peerpgid(const hgclient_t *hgc)
{
	assert(hgc);
	return hgc->pgid;
}

pid_t hgc_peerpid(const hgclient_t *hgc)
{
	assert(hgc);
	return hgc->pid;
}

/*!
 * Send command line arguments to let the server load the repo config and check
 * whether it can process our request directly or not.
 * Make sure hgc_setenv is called before calling this.
 *
 * @return - NULL, the server believes it can handle our request, or does not
 *           support "validate" command.
 *         - a list of strings, the server probably cannot handle our request
 *           and it sent instructions telling us what to do next. See
 *           chgserver.py for possible instruction formats.
 *           the list should be freed by the caller.
 *           the last string is guaranteed to be NULL.
 */
const char **hgc_validate(hgclient_t *hgc, const char *const args[],
			  size_t argsize)
{
	assert(hgc);
	if (!(hgc->capflags & CAP_VALIDATE))
		return NULL;

	packcmdargs(&hgc->ctx, args, argsize);
	writeblockrequest(hgc, "validate");
	handleresponse(hgc);

	/* the server returns '\0' if it can handle our request */
	if (hgc->ctx.datasize <= 1)
		return NULL;

	/* make sure the buffer is '\0' terminated */
	enlargecontext(&hgc->ctx, hgc->ctx.datasize + 1);
	hgc->ctx.data[hgc->ctx.datasize] = '\0';
	return unpackcmdargsnul(&hgc->ctx);
}

/*!
 * Execute the specified Mercurial command
 *
 * @return result code
 */
int hgc_runcommand(hgclient_t *hgc, const char *const args[], size_t argsize)
{
	assert(hgc);

	packcmdargs(&hgc->ctx, args, argsize);
	writeblockrequest(hgc, "runcommand");
	handleresponse(hgc);

	int32_t exitcode_n;
	if (hgc->ctx.datasize != sizeof(exitcode_n)) {
		abortmsg("unexpected size of exitcode");
	}
	memcpy(&exitcode_n, hgc->ctx.data, sizeof(exitcode_n));
	return ntohl(exitcode_n);
}

/*!
 * (Re-)send client's stdio channels so that the server can access to tty
 */
void hgc_attachio(hgclient_t *hgc)
{
	assert(hgc);
	if (!(hgc->capflags & CAP_ATTACHIO))
		return;
	attachio(hgc);
}

/*!
 * Update server's environment variables
 *
 * @param envp  list of environment variables in "NAME=VALUE" format,
 *              terminated by NULL.
 */
void hgc_setenv(hgclient_t *hgc, const char *const envp[])
{
	assert(hgc && envp);
	if (!(hgc->capflags & CAP_SETENV))
		return;
	packcmdargs(&hgc->ctx, envp, /*argsize*/ -1);
	writeblockrequest(hgc, "setenv");
}