reynir
/
albatross
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Albatrossd: --enable-stats and --retry-connections (#33) 

* Make albatrossd wait for other sockets to appear if --retry-connections is specified
* Connect to log/console before creating main socket
* Disable stats by default unless --enable stats is specified
This commit is contained in:
Magnus Skjegstad 2020-05-19 20:51:32 +02:00 committed by GitHub
parent 604f3ec52e
commit 50958a32f5
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 37 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
command-line/albatross_cli.ml
View File

@ -353,6 +353,14 @@ let exits =
remote_command_failed ::
Term.default_exits
let enable_stats =
let doc = "Connect to albatross-stats to report statistics" in
Arg.(value & flag & info [ "enable-stats" ] ~doc)
let retry_connections =
let doc = "Number of retries when connecting to other daemons (log, console, stats etc). 0 aborts after one failure, -1 is unlimited retries." in
Arg.(value & opt int 3 & info [ "retry-connections" ] ~doc)
let auth_exits =
[ Term.exit_info ~doc:"on local authentication failure \
(certificate not accepted by remote)"

39
daemon/albatrossd.ml
View File

@ -135,7 +135,7 @@ let write_reply name fd txt (hdr, cmd) =
let m = conn_metrics "unix"
let jump _ influx tmpdir dbdir =
let jump _ influx tmpdir dbdir retries enable_stats =
Sys.(set_signal sigpipe Signal_ignore);
Albatross_cli.set_tmpdir tmpdir;
Albatross_cli.set_dbdir dbdir;
@ -145,14 +145,37 @@ let jump _ influx tmpdir dbdir =
| Error (`Msg msg) -> Logs.err (fun m -> m "bailing out: %s" msg)
| Ok old_unikernels ->
Lwt_main.run
(let unix_connect s =
Vmm_lwt.connect Lwt_unix.PF_UNIX (Lwt_unix.ADDR_UNIX (socket_path s))
(let rec unix_connect ~retries s =
(Vmm_lwt.connect Lwt_unix.PF_UNIX (Lwt_unix.ADDR_UNIX (socket_path s)) >>= function
| Some x -> Lwt.return (Some x)
| None when (retries <> 0) ->
(Logs.err (fun m -> m "unable to connect to %s, retrying in 5 seconds" (socket_path s));
Lwt_unix.sleep 5.0 >>= fun () ->
unix_connect ~retries:(retries - 1) s)
| None -> Lwt.return_none)
in
init_influx "albatross" influx;
Vmm_lwt.server_socket `Vmmd >>= fun ss ->
(unix_connect `Log >|= function
(unix_connect ~retries `Log >|= function
| None -> invalid_arg "cannot connect to log socket"
| Some l -> l) >>= fun l ->
(unix_connect ~retries `Console >|= function
| None -> invalid_arg "cannot connect to console socket"
| Some c -> c) >>= fun c ->
(if enable_stats then
(unix_connect ~retries `Stats >|= function
| None -> invalid_arg "cannot connect to stats socket"
| Some c -> Some c)
else
Lwt.return_none)
>>= fun s ->
Lwt.catch (fun () ->
Vmm_lwt.server_socket `Vmmd)
(fun _ -> invalid_arg ("unable to create server socket " ^ (socket_path `Vmmd)))
>>= fun ss ->
let self_destruct_mutex = Lwt_mutex.create () in
let self_destruct () =
Lwt_mutex.with_lock self_destruct_mutex (fun () ->
@ -166,10 +189,6 @@ let jump _ influx tmpdir dbdir =
Vmm_lwt.safe_close ss)
in
Sys.(set_signal sigterm (Signal_handle (fun _ -> Lwt.async self_destruct)));
(unix_connect `Console >|= function
| None -> invalid_arg "cannot connect to console socket"
| Some c -> c) >>= fun c ->
unix_connect `Stats >>= fun s ->
let log_out txt wire = write_reply "log" l txt wire >|= fun _ -> ()
and cons_out = write_reply "cons" c
@ -201,7 +220,7 @@ let jump _ influx tmpdir dbdir =
open Cmdliner
let cmd =
Term.(const jump $ setup_log $ influx $ tmpdir $ dbdir),
Term.(const jump $ setup_log $ influx $ tmpdir $ dbdir $ retry_connections $ enable_stats),
Term.info "albatrossd" ~version:Albatross_cli.version
let () = match Term.eval cmd with `Ok () -> exit 0 | _ -> exit 1