Rooba
/
boinc
mirror of https://github.com/BOINC/boinc.git
1
0
Fork 0
Code Issues 1 Packages Projects Releases Wiki Activity
boinc/sched/handle_request.C

1066 lines
35 KiB
C
Raw Blame History

// Berkeley Open Infrastructure for Network Computing
// http://boinc.berkeley.edu
// Copyright (C) 2005 University of California
//
// This is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation;
// either version 2.1 of the License, or (at your option) any later version.
//
// This software is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
// See the GNU Lesser General Public License for more details.
//
// To view the GNU Lesser General Public License visit
// http://www.gnu.org/copyleft/lesser.html
// or write to the Free Software Foundation, Inc.,
// 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
// Handle a scheduling server RPC
#include <cassert>
#include <cstdio>
#include <vector>
#include <string>
#include <ctime>
#include <cmath>
using namespace std;
#include <unistd.h>
#include <sys/wait.h>
#include "boinc_db.h"
#include "backend_lib.h"
#include "error_numbers.h"
#include "parse.h"
#include "util.h"
#include "server_types.h"
#include "sched_util.h"
#include "main.h"
#include "handle_request.h"
#include "sched_msgs.h"
#include "sched_send.h"
#include "sched_config.h"
#include "../lib/filesys.h"
#ifdef _USING_FCGI_
#include "fcgi_stdio.h"
#endif
// Look up the host and its user, and make sure the authenticator matches.
// If no host ID is supplied, or if RPC seqno mismatch,
// create a new host record and return its ID
//
// POSTCONDITION:
// If this returns zero, then:
// - reply.host contains a valid host record (possibly new)
// - reply.user contains a valid user record
// - if user has team, reply.team contains team record
//
int authenticate_user(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
int retval;
char buf[256];
DB_HOST host;
DB_USER user;
DB_TEAM team;
if (sreq.hostid) {
retval = host.lookup_id(sreq.hostid);
if (retval) {
USER_MESSAGE um("Can't find host record", "low");
reply.insert_message(um);
log_messages.printf(
SCHED_MSG_LOG::NORMAL,
"[HOST#%d?] can't find host\n",
sreq.hostid
);
sreq.hostid = 0;
goto lookup_user_and_make_new_host;
}
reply.host = host;
strlcpy(
user.authenticator, sreq.authenticator,
sizeof(user.authenticator)
);
sprintf(buf, "where authenticator='%s'", user.authenticator);
retval = user.lookup(buf);
if (retval) {
USER_MESSAGE um("Invalid or missing account key. "
"Visit this project's web site to get an account key.",
"low"
);
reply.insert_message(um);
reply.request_delay = 3600;
reply.nucleus_only = true;
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[HOST#%d] [USER#%d] Bad authenticator '%s'\n",
host.id, user.id, sreq.authenticator
);
return ERR_AUTHENTICATOR;
}
reply.user = user;
if (host.userid != user.id) {
// If the request's host ID isn't consistent with the authenticator,
// create a new host record.
//
log_messages.printf(
SCHED_MSG_LOG::NORMAL,
"[HOST#%d] [USER#%d] inconsistent host ID; creating new host\n",
host.id, user.id
);
goto make_new_host;
}
// If the seqno from the host is less than what we expect,
// the user must have copied the state file to a different host.
// Make a new host record.
//
if (sreq.rpc_seqno < reply.host.rpc_seqno) {
sreq.hostid = 0;
log_messages.printf(
SCHED_MSG_LOG::NORMAL,
"[HOST#%d] [USER#%d] RPC seqno %d less than expected %d; creating new host\n",
reply.host.id, user.id, sreq.rpc_seqno, reply.host.rpc_seqno
);
goto make_new_host;
}
reply.host.rpc_seqno = sreq.rpc_seqno;
} else {
// here no hostid was given; we'll have to create a new host record
//
lookup_user_and_make_new_host:
strlcpy(
user.authenticator, sreq.authenticator,
sizeof(user.authenticator)
);
sprintf(buf, "where authenticator='%s'", user.authenticator);
retval = user.lookup(buf);
if (retval) {
USER_MESSAGE um(
"Invalid or missing account key. "
"Visit this project's web site to get an account key.",
"low"
);
reply.insert_message(um);
reply.request_delay = 3600;
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[HOST#<none>] Bad authenticator '%s'\n",
sreq.authenticator
);
return ERR_AUTHENTICATOR;
}
reply.user = user;
make_new_host:
// reply.user is filled in and valid at this point
//
host = sreq.host;
host.id = 0;
host.create_time = time(0);
host.userid = reply.user.id;
host.rpc_seqno = 0;
strcpy(host.venue, reply.user.venue);
host.fix_nans();
retval = host.insert();
if (retval) {
USER_MESSAGE um("Couldn't create host record in database", "low");
reply.insert_message(um);
boinc_db.print_error("host.insert()");
log_messages.printf(SCHED_MSG_LOG::CRITICAL, "host.insert() failed\n");
return retval;
}
host.id = boinc_db.insert_id();
reply.host = host;
reply.hostid = reply.host.id;
// this tells client to updates its host ID
}
// have user record in reply.user at this point
//
if (reply.user.teamid) {
retval = team.lookup_id(reply.user.teamid);
if (!retval) reply.team = team;
}
// compute email hash
//
md5_block(
(unsigned char*)reply.user.email_addr,
strlen(reply.user.email_addr),
reply.email_hash
);
// see if new cross-project ID (i.e., greater than the one we have)
//
if (strlen(sreq.cross_project_id)) {
if (strcmp(sreq.cross_project_id, reply.user.cross_project_id) > 0) {
strcpy(reply.user.cross_project_id, sreq.cross_project_id);
reply.update_user_record = true;
}
}
return 0;
}
// somewhat arbitrary formula for credit as a function of CPU time.
// Could also include terms for RAM size, network speed etc.
//
static void compute_credit_rating(HOST& host, SCHEDULER_REQUEST& sreq) {
double cobblestone_factor = 100;
host.credit_per_cpu_sec =
(fabs(host.p_fpops)/1e9 + fabs(host.p_iops)/1e9)
* cobblestone_factor / (2 * SECONDS_PER_DAY);
}
// modify host struct based on request.
// Copy all fields that are determined by the client.
//
static int modify_host_struct(SCHEDULER_REQUEST& sreq, HOST& host) {
host.timezone = sreq.host.timezone;
strncpy(host.domain_name, sreq.host.domain_name, sizeof(host.domain_name));
strncpy(host.serialnum, sreq.host.serialnum, sizeof(host.serialnum));
if (strcmp(host.last_ip_addr, sreq.host.last_ip_addr)) {
strncpy(host.last_ip_addr, sreq.host.last_ip_addr, sizeof(host.last_ip_addr));
} else {
host.nsame_ip_addr++;
}
host.on_frac = sreq.host.on_frac;
host.connected_frac = sreq.host.connected_frac;
host.active_frac = sreq.host.active_frac;
host.p_ncpus = sreq.host.p_ncpus;
strncpy(host.p_vendor, sreq.host.p_vendor, sizeof(host.p_vendor));
// unlikely this will change
strncpy(host.p_model, sreq.host.p_model, sizeof(host.p_model));
host.p_fpops = sreq.host.p_fpops;
host.p_iops = sreq.host.p_iops;
host.p_membw = sreq.host.p_membw;
host.p_calculated = sreq.host.p_calculated;
strncpy(host.os_name, sreq.host.os_name, sizeof(host.os_name));
strncpy(host.os_version, sreq.host.os_version, sizeof(host.os_version));
host.m_nbytes = sreq.host.m_nbytes;
host.m_cache = sreq.host.m_cache;
host.m_swap = sreq.host.m_swap;
host.d_total = sreq.host.d_total;
host.d_free = sreq.host.d_free;
host.n_bwup = sreq.host.n_bwup;
host.n_bwdown = sreq.host.n_bwdown;
host.fix_nans();
compute_credit_rating(host, sreq);
return 0;
}
static int update_host_record(HOST& xhost, USER& user) {
DB_HOST host;
int retval;
char buf[1024];
host = xhost;
if (strlen(host.host_cpid)) {
sprintf(buf, "%s%s", host.host_cpid, user.email_addr);
md5_block((const unsigned char*)buf, strlen(buf), host.host_cpid);
}
char* p = getenv("REMOTE_ADDR");
if (p) {
strlcpy(host.external_ip_addr, p, sizeof(host.external_ip_addr));
}
retval = host.update();
if (retval) {
log_messages.printf(SCHED_MSG_LOG::CRITICAL, "host.update() failed: %d\n", retval);
}
return 0;
}
// Decide which global prefs to use,
// (from request msg, or if absent then from user record)
// and parse them into the request message global_prefs field.
// Decide whether to send global prefs in reply msg
//
int handle_global_prefs(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
reply.send_global_prefs = false;
if (strlen(sreq.global_prefs_xml)) {
unsigned req_mod_time=0, db_mod_time=0;
bool same_account = !strcmp(
sreq.global_prefs_source_email_hash, reply.email_hash
);
bool update_prefs = false;
parse_int(sreq.global_prefs_xml, "<mod_time>", (int&)req_mod_time);
if (strlen(reply.user.global_prefs)) {
parse_int(reply.user.global_prefs, "<mod_time>", (int&)db_mod_time);
// if user record has more recent prefs,
// use them and arrange to return in reply msg
//
if (req_mod_time < db_mod_time) {
strcpy(sreq.global_prefs_xml, reply.user.global_prefs);
reply.send_global_prefs = true;
} else {
if (same_account) update_prefs = true;
}
} else {
if (same_account) update_prefs = true;
}
if (update_prefs) {
strcpy(reply.user.global_prefs, sreq.global_prefs_xml);
reply.update_user_record = true;
}
} else {
// request message has no global prefs;
// copy from user record, and send them in reply
//
if (strlen(reply.user.global_prefs)) {
strcpy(sreq.global_prefs_xml, reply.user.global_prefs);
reply.send_global_prefs = true;
}
}
sreq.global_prefs.parse(sreq.global_prefs_xml, reply.host.venue);
return 0;
}
// New handle completed results
//
int handle_results(
SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply
) {
DB_SCHED_RESULT_ITEM_SET result_handler;
SCHED_RESULT_ITEM* srip;
unsigned int i;
int retval;
RESULT* rp;
if (sreq.results.size() == 0) return 0;
// read all results the user is reporting
//
for (i=0; i<sreq.results.size(); i++) {
result_handler.add_result(sreq.results[i].name);
}
// read all results with the same name from the database into
// memory
//
retval = result_handler.enumerate();
if (retval) {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[HOST#%d] Batch query failed\n",
reply.host.id
);
}
// loop over all results that came back from the host
//
for (i=0; i<sreq.results.size(); i++) {
// a pointer to the current result from the host
//
rp = &sreq.results[i];
// acknowledge the result even if we couldn't find it --
// don't want it to keep coming back
//
reply.result_acks.push_back(*rp);
// get the result with the same name that came back from the
// database and point srip to it. Quantities that MUST be
// read from the DB are those where srip appears as an rval.
// These are: id, name, server_state, received_time, hostid.
// // Quantities that must be WRITTEN to the DB are those for
// which srip appears as an lval. These are:
// hostid,
// teamid, received_time, client_state, cpu_time, exit_status,
// app_version_num, claimed_credit, server_state, stderr_out,
// xml_doc_out, outcome, validate_state
retval = result_handler.lookup_result(rp->name, &srip);
if (retval) {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[HOST#%d] [RESULT#? %s] can't find result\n",
reply.host.id, rp->name
);
// no need to skip when updating DB, since it's not there!
continue;
}
log_messages.printf(
SCHED_MSG_LOG::NORMAL, "[HOST#%d] [RESULT#%d %s] got result\n",
reply.host.id, srip->id, srip->name
);
// Comment -- In the sanity checks that follow, should we
// verify that the results validate_state is consistent with
// this being a newly arrived result? What happens if a
// workunit was canceled after a result was sent? When it
// gets back in, do we want to leave the validate state 'as
// is'? Probably yes, which is as the code currently behaves.
if (srip->server_state == RESULT_SERVER_STATE_UNSENT) {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[HOST#%d] [RESULT#%d %s] got unexpected result: server state is %d\n",
reply.host.id, srip->id, srip->name, srip->server_state
);
srip->id=0; // mark to skip when updating DB
continue;
}
if (srip->received_time) {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[HOST#%d] [RESULT#%d %s] got result twice\n",
reply.host.id, srip->id, srip->name
);
srip->id=0; // mark to skip when updating DB
continue;
}
if (srip->hostid != sreq.hostid) {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[HOST#%d] [RESULT#%d %s] got result from wrong host; expected [HOST#%d]\n",
reply.host.id, srip->id, srip->name, srip->hostid
);
DB_HOST result_host;
int retval = result_host.lookup_id(srip->hostid);
if (retval) {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[RESULT#%d %s] Can't lookup [HOST#%d]\n",
srip->id, srip->name, srip->hostid
);
srip->id=0; // mark to skip when updating DB
continue;
} else if (result_host.userid != reply.host.userid) {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[USER#%d] [HOST#%d] [RESULT#%d %s] Not even the same user; expected [USER#%d]\n",
reply.host.userid, reply.host.id, srip->id, srip->name, result_host.userid
);
srip->id=0; // mark to skip when updating DB
continue;
} else {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[HOST#%d] [RESULT#%d %s] Allowing result because same USER#%d\n",
reply.host.id, srip->id, srip->name, reply.host.userid
);
}
} // hostids do not match
// modify the result record in the in-memory copy obtained
// from the DB earlier. If we found a problem above, we have
// continued and skipped this modify
//
srip->hostid = reply.host.id;
srip->teamid = reply.user.teamid;
srip->received_time = time(0);
srip->client_state = rp->client_state;
srip->cpu_time = rp->cpu_time;
srip->exit_status = rp->exit_status;
srip->app_version_num = rp->app_version_num;
srip->claimed_credit = rp->cpu_time * reply.host.credit_per_cpu_sec;
#if 1
log_messages.printf(SCHED_MSG_LOG::DEBUG,
"cpu %f cpcs %f, cc %f\n", srip->cpu_time, reply.host.credit_per_cpu_sec, srip->claimed_credit
);
#endif
srip->server_state = RESULT_SERVER_STATE_OVER;
safe_strncpy(srip->stderr_out, rp->stderr_out, sizeof(srip->stderr_out));
safe_strncpy(srip->xml_doc_out, rp->xml_doc_out, sizeof(srip->xml_doc_out));
// look for exit status and app version in stderr_out
// (historical - can be deleted at some point)
//
parse_int(srip->stderr_out, "<exit_status>", srip->exit_status);
parse_int(srip->stderr_out, "<app_version>", srip->app_version_num);
if ((srip->client_state == RESULT_FILES_UPLOADED) && (srip->exit_status == 0)) {
srip->outcome = RESULT_OUTCOME_SUCCESS;
log_messages.printf(SCHED_MSG_LOG::DEBUG,
"[RESULT#%d %s]: setting outcome SUCCESS\n",
srip->id, srip->name
);
} else {
log_messages.printf(SCHED_MSG_LOG::DEBUG,
"[RESULT#%d %s]: client_state %d exit_status %d; setting outcome ERROR\n",
srip->id, srip->name, srip->client_state, srip->exit_status
);
srip->outcome = RESULT_OUTCOME_CLIENT_ERROR;
srip->validate_state = VALIDATE_STATE_INVALID;
}
} // end of loop over all incoming results
// update all the results we have kept in memory, by storing to
// database.
//
if (config.use_transactions) {
retval = boinc_db.start_transaction();
if (retval) {
log_messages.printf(SCHED_MSG_LOG::CRITICAL,
"[HOST#%d] result_handler.start_transaction() == %d\n",
reply.host.id, retval
);
}
}
for (i=0; i<result_handler.results.size(); i++) {
// skip items that we previously marked to skip!
if (result_handler.results[i].id > 0) {
retval = result_handler.update_result(result_handler.results[i]);
if (retval) {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[HOST#%d] [RESULT#%d %s] can't update result: %s\n",
reply.host.id, result_handler.results[i].id, result_handler.results[i].name,
boinc_db.error_string()
);
}
}
}
// trigger the transition handle for the results' WUs
//
retval = result_handler.update_workunits();
if (retval) {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[HOST#%d] can't update WUs: %d\n",
reply.host.id, retval
);
}
if (config.use_transactions) {
retval = boinc_db.commit_transaction();
if (retval) {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[HOST#%d] result_handler.commit_transaction() == %d\n",
reply.host.id, retval
);
}
}
return 0;
}
// if the client has an old code sign public key,
// send it the new one, with a signature based on the old one.
// If they don't have a code sign key, send them one
//
void send_code_sign_key(
SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply, char* code_sign_key
) {
char* oldkey, *signature;
int i, retval;
char path[256];
if (strlen(sreq.code_sign_key)) {
if (strcmp(sreq.code_sign_key, code_sign_key)) {
log_messages.printf(SCHED_MSG_LOG::NORMAL, "received old code sign key\n");
// look for a signature file
//
for (i=0; ; i++) {
sprintf(path, "%s/old_key_%d", config.key_dir, i);
retval = read_file_malloc(path, oldkey);
if (retval) {
USER_MESSAGE um(
"You may have an outdated code verification key. "
"This may prevent you from accepting new executables. "
"If the problem persists, detach/attach the project. ",
"high"
);
reply.insert_message(um);
return;
}
if (!strcmp(oldkey, sreq.code_sign_key)) {
sprintf(path, "%s/signature_%d", config.key_dir, i);
retval = read_file_malloc(path, signature);
if (retval) {
USER_MESSAGE um(
"You may have an outdated code verification key. "
"This may prevent you from accepting new executables. "
"If the problem persists, detach/attach the project. ",
"high"
);
reply.insert_message(um);
} else {
safe_strcpy(reply.code_sign_key, code_sign_key);
safe_strcpy(reply.code_sign_key_signature, signature);
free(signature);
}
}
free(oldkey);
return;
}
}
} else {
safe_strcpy(reply.code_sign_key, code_sign_key);
}
}
bool wrong_core_client_version(
SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply
) {
char msg[256];
bool wrong_version = false;
if (sreq.core_client_major_version != BOINC_MAJOR_VERSION) {
// TODO: check for user-agent not empty and not BOINC
wrong_version = true;
sprintf(msg,
"To participate in this project, "
"you must use major version %d of the BOINC core client. "
"Your core client is major version %d.",
BOINC_MAJOR_VERSION,
sreq.core_client_major_version
);
log_messages.printf(
SCHED_MSG_LOG::NORMAL,
"[HOST#%d] [auth %s] Wrong major version from user: wanted %d, got %d\n",
sreq.hostid, sreq.authenticator,
BOINC_MAJOR_VERSION, sreq.core_client_major_version
);
} else if (config.min_core_client_version) {
int major = config.min_core_client_version/100;
int minor = config.min_core_client_version % 100;
if (sreq.core_client_minor_version < minor) {
wrong_version = true;
sprintf(msg,
"To participate in this project, "
"you must use version %d.02%d or higher of the BOINC core client. "
"Your core client is version %d.02%d.",
major, minor,
sreq.core_client_major_version, sreq.core_client_minor_version
);
log_messages.printf(
SCHED_MSG_LOG::NORMAL,
"[HOST#%d] [auth %s] Wrong minor version from user: wanted %d, got %d\n",
sreq.hostid, sreq.authenticator,
minor, sreq.core_client_minor_version
);
}
}
if (wrong_version) {
USER_MESSAGE um(msg, "low");
reply.insert_message(um);
reply.probable_user_browser = true;
reply.request_delay = 3600*24;
return true;
}
return false;
}
inline static const char* get_remote_addr() {
const char * r = getenv("REMOTE_ADDR");
return r ? r : "?.?.?.?";
}
void handle_msgs_from_host(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
unsigned int i;
DB_MSG_FROM_HOST mfh;
int retval;
for (i=0; i<sreq.msgs_from_host.size(); i++) {
reply.send_msg_ack = true;
MSG_FROM_HOST_DESC& md = sreq.msgs_from_host[i];
mfh.clear();
mfh.create_time = time(0);
safe_strcpy(mfh.variety, md.variety);
mfh.hostid = reply.host.id;
mfh.handled = false;
safe_strcpy(mfh.xml, md.msg_text.c_str());
log_messages.printf(
SCHED_MSG_LOG::NORMAL,
"got msg from host; variety %s text %s\n",
mfh.variety, mfh.xml
);
retval = mfh.insert();
if (retval) {
log_messages.printf(SCHED_MSG_LOG::CRITICAL,
"[HOST#%d] message insert failed: %d\n",
reply.host.id, retval
);
}
}
}
void handle_msgs_to_host(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
DB_MSG_TO_HOST mth;
char buf[256];
sprintf(buf, "where hostid = %d and handled = %d", reply.host.id, 0);
while (!mth.enumerate(buf)) {
reply.msgs_to_host.push_back(mth);
mth.handled = true;
mth.update();
}
}
#if 0
void notify_if_newer_core_version(
SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply,
PLATFORM& platform, SCHED_SHMEM& ss
) {
CORE_VERSION* cvp;
int req_version;
if (sreq.core_client_major_version != BOINC_MAJOR_VERSION) {
return;
}
req_version = sreq.core_client_major_version*100 + sreq.core_client_minor_version;
cvp = ss.lookup_core_version(platform.id);
if (cvp && cvp->version_num > req_version) {
sprintf(reply.message,
"A new version (%d.%02d) of the BOINC core client is available from this project's web site.",
cvp->version_num/100,
cvp->version_num%100
);
strcpy(reply.message_priority, "low");
}
}
#endif
void process_request(
SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply, SCHED_SHMEM& ss,
char* code_sign_key
) {
PLATFORM* platform;
int retval;
double last_rpc_time;
struct tm *rpc_time_tm;
int last_rpc_dayofyear;
int current_rpc_dayofyear;
bool ok_to_send_work = true;
bool have_no_work;
char buf[256];
// if different major version of BOINC, just send a message
//
if (wrong_core_client_version(sreq, reply)) {
ok_to_send_work = false;
// if no results, return without accessing DB
//
if (sreq.results.size() == 0) {
return;
}
}
// if there's no work, and client isn't returning results,
// this isn't an initial RPC,
// and client is requesting work, return without accessing DB
//
lock_sema();
have_no_work = ss.no_work(g_pid);
unlock_sema();
if ((sreq.work_req_seconds > 0)
&& have_no_work
&& (sreq.results.size() == 0)
&& (sreq.hostid != 0)
) {
USER_MESSAGE um("No work available", "low");
reply.insert_message(um);
reply.request_delay = 3600;
if (!config.msg_to_host) {
log_messages.printf(
SCHED_MSG_LOG::NORMAL, "No work - skipping DB access\n"
);
return;
}
}
// FROM HERE ON DON'T RETURN; goto leave instead
// because we've tagged an entry in the work array with our process ID
// now open the database
//
retval = open_database();
if (retval) {
send_message("Server can't open database", 3600);
goto leave;
}
retval = authenticate_user(sreq, reply);
if (retval) goto leave;
if (reply.user.id == 0) {
log_messages.printf(SCHED_MSG_LOG::CRITICAL, "No user ID!\n");
}
log_messages.printf(
SCHED_MSG_LOG::NORMAL,
"Processing request from [USER#%d] [HOST#%d] [IP %s] [RPC#%d] core client version %d.%02d\n",
reply.user.id, reply.host.id, get_remote_addr(), sreq.rpc_seqno,
sreq.core_client_major_version, sreq.core_client_minor_version
);
++log_messages;
last_rpc_time = reply.host.rpc_time;
rpc_time_tm = localtime((const time_t*)&reply.host.rpc_time);
last_rpc_dayofyear = rpc_time_tm->tm_yday;
reply.host.rpc_time = time(0);
rpc_time_tm = localtime((const time_t*)&reply.host.rpc_time);
current_rpc_dayofyear = rpc_time_tm->tm_yday;
if (last_rpc_dayofyear != current_rpc_dayofyear) {
log_messages.printf(SCHED_MSG_LOG::DEBUG, "[HOST#%d] Resetting nresults_today\n", reply.host.id);
reply.host.nresults_today = 0;
}
retval = modify_host_struct(sreq, reply.host);
// look up the client's platform in the DB
//
platform = ss.lookup_platform(sreq.platform_name);
if (!platform) {
sprintf(buf, "platform '%s' not found", sreq.platform_name);
USER_MESSAGE um(buf, "low");
reply.insert_message(um);
log_messages.printf(
SCHED_MSG_LOG::CRITICAL, "[HOST#%d] platform '%s' not found\n",
reply.host.id, sreq.platform_name
);
reply.request_delay = 3600*24;
goto leave;
}
#if 0
notify_if_newer_core_version(sreq, reply, *platform, ss);
#endif
handle_global_prefs(sreq, reply);
#if 0
reply.deletion_policy_priority = config.deletion_policy_priority;
reply.deletion_policy_expire = config.deletion_policy_expire;
#endif
if (reply.update_user_record) {
DB_USER user;
user = reply.user;
user.update();
}
handle_results(sreq, reply);
// if last RPC was within config.min_sendwork_interval, don't send work
//
if (!have_no_work && ok_to_send_work && sreq.work_req_seconds > 0) {
if (config.min_sendwork_interval) {
double diff = dtime() - last_rpc_time;
if (diff < config.min_sendwork_interval) {
ok_to_send_work = false;
log_messages.printf(
SCHED_MSG_LOG::NORMAL,
"Not sending work - last RPC too recent: %f\n", diff
);
sprintf(buf,
"Not sending work - last RPC too recent: %d sec", (int)diff
);
USER_MESSAGE um(buf, "low");
reply.insert_message(um);
reply.request_delay = config.min_sendwork_interval;
}
}
if (ok_to_send_work) {
send_work(sreq, reply, *platform, ss);
}
}
send_code_sign_key(sreq, reply, code_sign_key);
handle_msgs_from_host(sreq, reply);
if (config.msg_to_host) {
handle_msgs_to_host(sreq, reply);
}
update_host_record(reply.host, reply.user);
leave:
if (!have_no_work) {
ss.restore_work(g_pid);
}
}
extern double max_allowable_disk(SCHEDULER_REQUEST& req);
extern double watch_diskspace[3];
// returns zero if there is a file we can delete. Return non-zero if
// no file to delete.
//
int delete_file_from_host(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& sreply) {
int nfiles = (int)sreq.file_infos.size();
char buf[256];
if (!nfiles) {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"[HOST#%d]: no disk space but no files we can delete!\n", sreply.host.id
);
sprintf(buf,
"No disk space (BOINC needs %.1f MB more). ",
max_allowable_disk(sreq)/1.e6
);
if (watch_diskspace[0] != 0.0) {
strcat(buf, "Review preferences for maximum disk space used.");
} else if (watch_diskspace[1] != 0.0) {
strcat(buf, "Review preferences for maximum disk percentage used.");
} else if (watch_diskspace[2] != 0.0) {
strcat(buf, "Review preferences for minimum disk free space allowed.");
}
USER_MESSAGE um(buf, "high");
sreply.insert_message(um);
sreply.request_delay = 24*3600;
return 1;
}
// pick a data file to delete. Do this deterministically
// so that we always tell host to delete the same file. But to prevent
// all hosts from removing 'the same' file, we choose a file which depends
// upon the hostid.
//
// Assumption is that if nothing has changed on the host, the order in
// which it reports files is fixed. If this is false, we need to sort
// files into order by name!
//
int j = sreply.host.id % nfiles;
FILE_INFO& fi = sreq.file_infos[j];
sreply.file_deletes.push_back(fi);
log_messages.printf(
SCHED_MSG_LOG::DEBUG,
"[HOST#%d]: delete file %s (make space)\n", sreply.host.id, fi.name
);
// give host 4 hours to nuke the file and come back. This might
// in general be too soon, since host needs to complete any work
// that depends upon this file, before it will be removed by core client.
//
sprintf(buf, "Removing file %s to free up disk space", fi.name);
USER_MESSAGE um(buf, "low");
sreply.insert_message(um);
sreply.request_delay = 4*3600;
return 0;
}
void debug_sched(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& sreply, const char *trigger) {
char tmpfilename[256];
FILE *fp;
if (!boinc_file_exists(trigger)) {
return;
}
sprintf(tmpfilename, "sched_reply_%06d_%06d", sreq.hostid, sreq.rpc_seqno);
// use _XXXXXX if you want random filenames rather than
// deterministic mkstemp(tmpfilename);
fp=fopen(tmpfilename, "w");
if (!fp) {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"Found %s, but can't open %s\n", trigger, tmpfilename
);
return;
}
log_messages.printf(
SCHED_MSG_LOG::DEBUG,
"Found %s, so writing %s\n", trigger, tmpfilename
);
sreply.write(fp);
fclose(fp);
sprintf(tmpfilename, "sched_request_%06d_%06d", sreq.hostid, sreq.rpc_seqno);
fp=fopen(tmpfilename, "w");
if (!fp) {
log_messages.printf(
SCHED_MSG_LOG::CRITICAL,
"Found %s, but can't open %s\n", trigger, tmpfilename
);
return;
}
log_messages.printf(
SCHED_MSG_LOG::DEBUG,
"Found %s, so writing %s\n", trigger, tmpfilename
);
sreq.write(fp);
fclose(fp);
return;
}
void handle_request(
FILE* fin, FILE* fout, SCHED_SHMEM& ss, char* code_sign_key
) {
SCHEDULER_REQUEST sreq;
SCHEDULER_REPLY sreply;
memset(&sreq, 0, sizeof(sreq));
if (sreq.parse(fin) == 0){
log_messages.printf(
SCHED_MSG_LOG::NORMAL,
"Handling request: IP %s, auth %s, host %d, platform %s, version %d.%02d\n",
get_remote_addr(), sreq.authenticator, sreq.hostid, sreq.platform_name,
sreq.core_client_major_version, sreq.core_client_minor_version
);
process_request(sreq, sreply, ss, code_sign_key);
} else {
log_messages.printf(
SCHED_MSG_LOG::NORMAL,
"Incomplete request received from IP %s, auth %s, platform %s, version %d.%02d\n",
get_remote_addr(), sreq.authenticator, sreq.platform_name,
sreq.core_client_major_version, sreq.core_client_minor_version
);
USER_MESSAGE um("Incomplete request received.", "low");
sreply.insert_message(um);
sreply.nucleus_only = true;
}
// if we got no work, and we have no file space, delete some files
if (sreply.results.size()==0 && (sreply.wreq.insufficient_disk || sreply.wreq.disk_available<0)) {
// try to delete a file to make more space. Also give some
// hints to the user about what's going wrong (lack of disk
// space).
delete_file_from_host(sreq, sreply);
}
#if 1
// You can call debug_sched() for whatever situation is of
// interest to you. It won't do anything unless you create
// (touch) the file 'debug_sched' in the project root directory.
//
if (sreply.results.size()==0 && sreply.hostid && sreq.work_req_seconds>1.0)
debug_sched(sreq, sreply, "../debug_sched");
else if (max_allowable_disk(sreq)<0 || (sreply.wreq.insufficient_disk || sreply.wreq.disk_available<0))
debug_sched(sreq, sreply, "../debug_sched");
#endif
sreply.write(fout);
}
const char *BOINC_RCSID_2ac231f9de = "$Id$";
Reference in New Issue View Git Blame Copy Permalink