зеркало из https://github.com/mozilla/gecko-dev.git
start storing the query string in the database
This commit is contained in:
Родитель
a19f47a6ab
Коммит
482b413f4c
|
@ -38,8 +38,6 @@ chdir $root_dir
|
|||
#utime($read_time, $end_time, $timestamp_file)
|
||||
# or die "Can't update timestamp on $timestamp_file: $!";
|
||||
|
||||
#CREATE TABLE entries (id INT PRIMARY KEY, protocol VARCHAR(4), protocol_version VARCHAR(5), client VARCHAR(15), date_time DATETIME, method VARCHAR(4), file_id INT, status CHAR(3), bytes INT, site_id TINYINT, log_id INT);
|
||||
|
||||
# Regular expressions that grab data from the log entries; pre-defined
|
||||
# and pre-compiled here for performance. The backslash in [^\"]
|
||||
# isn't necessary for Perl but fixes indenting confusion in emacs.
|
||||
|
@ -62,14 +60,17 @@ my $dbh = DBI->connect($dsn,
|
|||
# Prepare the statements we're going to use to insert HTTP log entries into
|
||||
# the database.
|
||||
my $insert_entry_sth = $dbh->prepare("INSERT INTO entries (id, protocol, protocol_version,
|
||||
client, date_time, method, file_id, status, bytes, site_id, log_id)
|
||||
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)");
|
||||
client, date_time, method, file_id, status, bytes,
|
||||
query_id, site_id, log_id)
|
||||
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)");
|
||||
my $insert_file_sth = $dbh->prepare("INSERT INTO files (id, path, name) VALUES (?, ?, ?)");
|
||||
my $get_file_id_sth = $dbh->prepare("SELECT id FROM files WHERE path = ? AND name = ?");
|
||||
my $get_log_status_sth = $dbh->prepare("SELECT id, status FROM logs WHERE path = ? AND name = ?");
|
||||
my $insert_log_sth = $dbh->prepare("INSERT INTO logs (id, path, name, site_id, status) VALUES (?, ?, ?, ?, ?)");
|
||||
my $update_log_sth = $dbh->prepare("UPDATE logs SET status = ? WHERE id = ?");
|
||||
my $get_site_id_sth = $dbh->prepare("SELECT id FROM sites WHERE abbr = ?");
|
||||
my $get_query_id_sth = $dbh->prepare("SELECT id FROM queries WHERE query = ?");
|
||||
my $insert_query_sth = $dbh->prepare("INSERT INTO queries (id, query) VALUES (?, ?)");
|
||||
|
||||
# Get the last unique ID from the database so we know what the next one
|
||||
# should be. XXX These assume only one script process will ever be running
|
||||
|
@ -80,6 +81,7 @@ my $get_site_id_sth = $dbh->prepare("SELECT id FROM sites WHERE abbr = ?");
|
|||
# a second process access the database at all.
|
||||
my ($entry_id) = $dbh->selectrow_array("SELECT MAX(id) FROM entries") || 0;
|
||||
my ($max_file_id) = $dbh->selectrow_array("SELECT MAX(id) FROM files") || 0;
|
||||
my ($max_query_id) = $dbh->selectrow_array("SELECT MAX(id) FROM queries") || 0;
|
||||
|
||||
my $seen = 0;
|
||||
my $entered = 0;
|
||||
|
@ -207,7 +209,16 @@ sub process_log {
|
|||
next unless $status == 200 || $status == 206;
|
||||
|
||||
# Strip the URL query string, if any, from the file string.
|
||||
$file = (split(/\?/, $file))[0];
|
||||
($file, my $query) = (split(/\?/, $file));
|
||||
|
||||
my $query_id;
|
||||
if ($query) {
|
||||
($query_id) = $dbh->selectrow_array($get_query_id_sth, {}, $query);
|
||||
if (!$query_id) {
|
||||
$query_id = ++$max_query_id;
|
||||
$insert_query_sth->execute($query_id, $query);
|
||||
}
|
||||
}
|
||||
|
||||
# Split up the file string into a path and a name.
|
||||
$file =~ /^(.*)\/([^\/]*)$/;
|
||||
|
@ -262,7 +273,7 @@ sub process_log {
|
|||
++$log_entered;
|
||||
$insert_entry_sth->execute($entry_id, $protocol, $protocol_version, $host,
|
||||
$date_time, $method, $file_id, $status, $bytes,
|
||||
$site_id, $log_id);
|
||||
$query_id, $site_id, $log_id);
|
||||
}
|
||||
close(LOGFILE);
|
||||
$update_log_sth->execute("processed", $log_id) || die $dbh->errstr;
|
||||
|
|
Загрузка…
Ссылка в новой задаче