Perlbal::Manual::Internals(3) Perlbal's architecture at a glance

VERSION

Perlbal 1.78.

DESCRIPTION

Connections come in from wherever and get to the TCPListener. It uses Service objects to determine what kind of Client* to spawn. The Client classes then handle crafting the response for the user.

                            {{ INTERNET }}
                                  |
                                  v
              [Service]<===>[TCPListener]
                          ___/    |    \___
                         v        v        v
             [ClientManage]  [ClientHTTP] [ClientProxy]
                                                ^
                                                |
                                                v
                                          [BackendHTTP]

Perlbal decides what backend to send a request to randomly (only presently supported method). If that service has idle backend connections available, configured by "backend_persist" and "connect_ahead", it will reuse those connections and greatly reduce latency. See more detail in Perlbal::Manual::LoadBalancer.

Perlbal also specializes in ``spoonfeeding'' data to slow clients. This allows backends to continue serving requests while Perlbal transfers responses back as fast as the client can read.

Classes

The following is a brief introduction/overview to the main Perlbal's classes:

Perlbal::Socket

Descends from Danga::Socket.

Adds on to the base class to provide some functionality specifically useful for creating HTTP sockets.

Fields

headers_string
Headers as they're being read.
req_headers
The final Perlbal::HTTPHeaders object inbound.
res_headers
Response headers outbound (Perlbal::HTTPHeaders object).
create_time
Creation time.
alive_time
Last time noted alive.
state
General purpose state; used by descendants.
do_die
If on, die and do no further requests.
read_buf
Arrayref of scalarref read from client.
read_ahead
Bytes sitting in read_buf.
read_size
Total bytes read from client, ever.
ditch_leading_rn
If true, the next header parsing will ignore a leading \r\n.
observed_ip_string
If defined, contains the observed IP string of the peer we're serving. This is intended for holding the value of the X-Forwarded-For and using it to govern ACLs.

Perlbal::TCPListener

Descends from Perlbal::Socket.

Very lightweight and fast connection accept class. Takes incoming connections as fast as possible and passes them off, instantiating one of the various Client* classes to handle it.

Fields

service
Perlbal::Service.
hostport
Scalar IP port of where this service is listening for new connections.
sslopts
The SSL Options.

    use Data::Dumper;
    warn Dumper( $tcp_listener->{'sslopts'} );

The above lines would print something like the following:

    $VAR1 = {
              'ssl' => {
                         'SSL_cipher_list' => '...',
                         'SSL_cert_file' => '...',
                         'SSL_key_file' => ',,,',
                         'SSL_ca_path' => '...',
                         'SSL_verify_mode' => '...'
                       }
            };
v6
Boolean value stating whether the installation of Perlbal supports IPv6 (which basically boils down to Danga::Socket v1.6.1 and IO::Socket::INET6 being available).

Perlbal::BackendHTTP

Descends from Perlbal::Socket.

This class handles connections to the backend web nodes for getting data back to the user. This class is used by other classes such as Perlbal::ClientProxy to send a request to an internal node.

Fields

client
Perlbal::ClientProxy connection, or undef.
service
Perlbal::Service.
pool
Perlbal::Pool; whatever pool we spawned from.
ip
IP scalar.
port
Port scalar.
ipport
"$ip:$port".
reportto
Object; must implement reporter interface.
has_attention
Has been accepted by a webserver and we know for sure we're not just talking to the TCP stack.
waiting_options
If true, we're waiting for an OPTIONS * response to determine when we have attention.
disconnect_at
Time this connection will be disconnected, if it's kept-alive and backend told us; otherwise "undef" for unknown.
content_length
Length of document being transferred. Only applies when the backend server sends a content-length header.
content_length_remain
Bytes remaining to be read. Only applies when the backend server sends a content-length header.
use_count
Number of requests this backend's been used for.
generation
Int; counts what generation we were spawned in.
buffered_upload_mode
Boolean. If on, we're doing a buffered upload transmit.
scratch
Extra storage; plugins can use it if they want.

Perlbal::HTTPHeaders

Header management. Parses headers (request and response) and stores data for further user. Also manages validation of the request line so that it conforms to HTTP specifications.

Fields

headers
href; lowercase header -> comma-sep list of values.
origcase
Href; lowercase header -> provided case.
hdorder
Aref; order headers were received (canonical order).
method
Scalar; request method (if GET request).
uri
Scalar; request URI (if GET request).
type
"res" or "req".
code
HTTP response status code.
codetext
Status text that for response code.
ver
Version (string) ``1.1''.
vernum
Version (number: major*1000+minor): ``1.1'' => 1001.
responseLine
First line of HTTP response (if response).
requestLine
First line of HTTP request (if request).

Perlbal::ClientHTTPBase

Descends from Perlbal::Socket.

Provides base functionality to Perlbal::ClientHTTP and Perlbal::ClientProxy. Notably, the ability to efficiently send files to the remote user. Also handles most of the state logic for statistics and such. Is also used for services of type "selector". Perlbal::ClientHTTPBase then reads in the request headers, and asks the service to re-bless the client instance to a more specific type, for either a Perlbal::ClientProxy or Perlbal::ClientHTTP (depending on selector's mapping).

Fields

service
Perlbal::Service object.
replacement_uri
URI to send instead of the one requested; this is used to instruct "_serve_request" to send an index file instead of trying to serve a directory and failing.
scratch
Extra storage; plugins can use it if they want.
reproxy_file
Filename the backend told us to start opening.
reproxy_file_size
Size of file, once we "stat()" it.
reproxy_fh
If needed, IO::Handle of fd.
reproxy_file_offset
How much we've sent from the file.
post_sendfile_cb
Subref to run after we're done sendfile'ing the current file.
requests
Number of requests this object has performed for the user.
selector_svc
The original service from which we came.
is_ssl
Whether the socket was SSL attached (restricted operations).

Perlbal::ClientHTTP

Descends from Perlbal::ClientHTTPBase.

Very simple and lightweight class. Handles sending files to the user without much overhead. Most of the functionality is contained in the parent class, and this class doesn't implement much new stuff.

Fields

put_in_progress
1 when we're currently waiting for an async job to return.
put_fh
File handle to use for writing data.
put_fh_filename
Filename of put_fh.
put_pos
File offset to write next data at.
content_length
Length of document being transferred.
content_length_remain
Bytes remaining to be read.
chunked_upload_state
Boolean/obj: if processing a chunked upload, Perlbal::ChunkedUploadState object, else undef.

Perlbal::ClientProxy

Descends from Perlbal::ClientHTTPBase.

Takes an incoming connection from a user and connects to a backend node ("Perlbal::BackendHTTP") and relays the request. The backend can then either tell the proxy to reproxy and load a file from disk, or return a file directly, or just return a status message.

Fields

backend
Perlbal::BackendHTTP object (or "undef" if disconnected).
backend_requested
True if we've requested a backend for this request.
reconnect_count
Number of times we've tried to reconnect to backend.
high_priority
Boolean; 1 if we are or were in the high priority queue.
low_priority
Boolean; 1 if we are or were in the low priority queue.
reproxy_uris
Arrayref; URIs to reproxy to, in order.
reproxy_expected_size
Int: size of response we expect to get back for reproxy.
currently_reproxying
Arrayref; the host info and URI we're reproxying right now.
content_length_remain
Int: amount of data we're still waiting for.
responded
Bool: whether we've already sent a response to the user or not.
last_request_time
Int: time that we last received a request.
primary_res_hdrs
If defined, we are doing a transparent reproxy-URI and the headers we get back aren't necessarily the ones we want. Instead, get most headers from the provided "res" headers object here.
is_buffering
Bool; if we're buffering some/all of a request to memory/disk.
is_writing
Bool; if on, we currently have an "aio_write" out.
start_time
Hi-res time when we started getting data to upload.
bufh
Buffered upload filehandle object.
bufilename
String; buffered upload filename.
bureason
String; if defined, the reason we're buffering to disk.
buoutpos
Int; buffered output position.
backend_stalled
Boolean: if backend has shut off its reads because we're too slow.
unread_data_waiting
Boolean: if we shut off reads while we know data is yet to be read from client.
chunked_upload_state
Bool/obj: if processing a chunked upload, Perlbal::ChunkedUploadState object, else undef.
request_body_length
Integer: request's body length, either as-declared, or calculated after chunked upload is complete.
last_upload_packet
Unixtime we last sent a UDP upload packet. For perlbal sending out UDP packets related to upload status (for xmlhttprequest upload bar).
upload_session
Client's self-generated upload session. For perlbal sending out UDP packets related to upload status (for xmlhttprequest upload bar).
retry_count
Number of times we've retried this request so far after getting 500 errors.

Perlbal::ClientManage

Descends from Perlbal::Socket.

Simple interface that provides a way for users to use the management interface of Perlbal. You can connect to the management port (as defined in the config file) with a web browser or regular telnet (see Perlbal::Manual::Management for more information on this).

Fields

service
Perlbal::Service.
buf
Read buffer.
is_http
Boolean stating whether the request is HTTP.
ctx
Perlbal::CommandContext.

Perlbal::Service

A service is a particular item that Perlbal is doing. Services can have a role which defines how they behave. Each service can also have a bunch of parameters set to further adjust its behavior. By itself, the Service class handles maintaining pools of backend connections and managing statistics about itself.

Fields

name
Name of the service.
role
Role type ("web_server", "reverse_proxy", etc).
enabled
Boolean; whether we're enabled or not (enabled = listening).
pool
Perlbal::Pool that we're using to allocate nodes if we're in proxy mode.
listener
Perlbal::TCPListener object, when enabled.
reproxy_cache
Perlbal::Cache object, when enabled.

End-user tunables

listen
"IP:port" of where we're listening for new connections.
docroot
Document root for "web_server" role.
dirindexing
Boolean; directory indexing (for "web_server" role). Not async.
index_files
Arrayref of filenames to try for index files.
enable_concatenate_get
Boolean; if user can request concatenated files.
enable_put
Boolean; whether PUT is supported.
max_put_size
Max size in bytes of a put file.
max_chunked_request_size
Max size in bytes of a chunked request (to be written to disk first).
min_put_directory
Number of directories required to exist at beginning of URIs in put.
enable_delete
Boolean; whether DELETE is supported.
high_priority_cookie
Cookie name to check if the client's requests should be considered high priority.

See also "high_priority_cookie_contents".

high_priority_cookie_contents
Aforementioned cookie value must contain this substring.
backend_persist_cache
Max number of persistent backends to hold onto while no clients.
persist_client
Boolean; persistent connections for clients.
persist_backend
Boolean; persistent connections for backends.
verify_backend
Boolean; get attention of backend before giving it clients (using OPTIONS).
verify_backend_path
Path to check with the OPTIONS request (default is "*").
max_backend_uses
Max requests to send per kept-alive backend (default 0 = unlimited).
connect_ahead
Number of spare backends to connect to in advance all the time.
buffer_size
How much data a Perlbal::ClientProxy object should buffer from a backend.
buffer_size_reproxy_url
Same as above but for backends that are reproxying for us.
queue_relief_size
Number of outstanding standard priority connections to activate pressure relief at.
queue_relief_chance
Int, 0-100; % chance to take a standard priority request when we're in pressure relief mode.
trusted_upstream_proxies
Array of Net::Netmask objects containing netmasks for trusted upstreams.
always_trusted
Boolean; if true, always trust upstreams.
blind_proxy
Boolean; if true, do not modify "X-Forwarded-For", "X-Host", or "X-Forwarded-Host" headers.
enable_reproxy
Boolean; if true, advertise that server will reproxy files and/or URLs.
reproxy_cache_maxsize
Maximum number of reproxy results to be cached. (0 is disabled and default).
client_sndbuf_size
Bytes for "SO_SNDBUF".
server_process
Path to server process (executable).
persist_client_idle_timeout
Keep-alive timeout in seconds for clients (default is 30).
idle_timeout
Idle timeout outside of keep-alive time (default is 30).

Internal state

waiting_clients
Arrayref of clients waiting for backendhttp connections.
waiting_clients_highpri
Arrayref of high-priority clients waiting for backendhttp connections.
waiting_clients_lowpri
Arrayref of low-priority clients waiting for backendhttp connections.
waiting_client_count
Number of clients waiting for backends.
waiting_client_map
Map of clientproxy fd -> 1 (if they're waiting for a connection).
pending_connects
Hashref of "ip:port" -> $time (only one pending connect to backend at a time).
pending_connect_count
Number of outstanding backend connects.
bored_backends
Arrayref of backends we've already connected to, but haven't got clients.
hooks
Hashref: hookname => [ [ plugin, ref ], [ plugin, ref ], ... ].
plugins
Hashref: name => 1.
plugin_order
Arrayref: name, name, name...
plugin_setters
Hashref: { plugin_name => { key_name => coderef } }.
extra_config
Hashref with extra config options; name => values.
spawn_lock
Boolean; if true, we're currently in "spawn_backends".
extra_headers
{ insert => [ [ header, value ], ... ], remove => [ header, header, ... ], set => [ [ header, value ], ... ] }.

Used in header management interface.

generation
Int; generation count so we can slough off backends from old pools.
backend_no_spawn
{ ``ip:port'' => 1 }.

If on, "spawn_backends" will ignore this "ip:port" combo.

buffer_backend_connect
0 if off; otherwise, number of bytes to buffer before we ask for a backend.
selector
CODE ref, or undef, for role "selector" services.
default_service
Name of a service a selector should default to.
buffer_uploads
Boolean; enable/disable the buffered uploads to disk system.
buffer_uploads_path
Path to store buffered upload files.
buffer_upload_threshold_time
Int; buffer uploads estimated to take longer than this.
buffer_upload_threshold_size
Int; buffer uploads greater than this size (in bytes).
buffer_upload_threshold_rate
Int; buffer uploads uploading at less than this rate (in bytes/sec).
upload_status_listeners
Comma separated list of "ip:port" of UDP upload status receivers.
upload_status_listeners_sockaddr
Arrayref of sockaddrs (packed ip/port).
enable_ssl
Boolean; whether this service speaks SSL to the client.
ssl_key_file
File path to key pem file.
ssl_cert_file
File to path to cert pem file.
ssl_cipher_list
OpenSSL cipher list string.
ssl_ca_path
Path to certificates directory.
ssl_verify_mode
Int; verification mode, see IO::Socket::SSL.
enable_error_retries
Boolean; whether we should retry requests after errors.
error_retry_schedule
Comma-separated seconds (full or partial) to delay between retries.
latency
Milliseconds of latency to add to request.
server_tokens
Boolean; whether to provide a ``Server'' header.
_stat_requests
Total requests to this service.
_stat_cache_hits
Total requests to this service that were served via the reproxy-url cache.