health.prg

<?php
// vim: tabstop=2 expandtab syn=php
/*
 * Perform basic sanity tests here to ensure this instance is healthy.
 * Returning a non 200 result (throwing an exception) will cause the
 * cluster to deprecate this node.
 */


require_once 'hcuEnv.i';


function check_database($env) {
    $db = $env['db'];
    $connect_to = sprintf('host=%s port=%d dbname=%s', $db['host'],
                          $db['port'], $db['dbname']);
    if (!is_null($db['user'])) {
        $connect_to .= ' user=' . $db['user'];
    }
    if (!is_null($db['connect_timeout'])) {
        $connect_to .= ' connect_timeout=' . $db['connect_timeout'];
    }
    if (!is_null($db['password'])) {
        $connect_to .= ' password=' . $db['password'];
    }
    $r = @pg_connect($connect_to);
    if (!$r) {
        throw new Exception('DB connect error');
    }
    if (pg_connection_status($r) !== PGSQL_CONNECTION_OK) {
        throw new Exception('DB status error');
    }
    if (!pg_ping($r)) {
        throw new Exception('DB ping error');
    }
}


function check_disk_usage($env) {
    $cwd_free = disk_free_space(".");
    $root_free = disk_free_space("/");
    if ($cwd_free < 1024 * 1024 * 500 or
        $root_free < 1024 * 1024 * 500) {
        throw new Exception('Disk(s) too full');
    }
    return [
        "cwd_free" => $cwd_free,
        "root_free" => $root_free
    ];
}

function check_mem_usage($env) {
    $mem = memory_get_usage(TRUE);
    if ($mem > (1024 * 1024 * 100)) {
        throw new Exception('Memory usage too high');
    }
    return [
        "usage" => $mem
    ];
}

function wrap_check($checkfunc, $env) {
    $res = [];
    try {
        return [TRUE, $checkfunc($env)];
    } catch (Exception $e) {
        return [FALSE, $e];
    }
}

function run_checks() {
    $env = LoadSystemEnv('health');

    $logger = $env['logger'];
    $checks = [
        "check_disk_usage",
        "check_mem_usage",
        "check_database"
    ];
    $results = [];
    $start_ts = microtime(TRUE);
    $overall_success = TRUE;
    foreach ($checks as $fn) {
        $check_ts = microtime(TRUE);
        list($success, $info_or_exc) = wrap_check($fn, $env);
        $time = microtime(TRUE) - $start_ts;
        $results[$fn] = ["time" => $time];
        if ($success) {
            $results[$fn]["success"] = TRUE;
            $results[$fn]["info"] = $info_or_exc;
        } else {
            if (!$env['devmode']) {
                http_response_code(500);
            } else {
              $logger->warning("DEVMODE preventing negative health response");
            }
            $msg = $info_or_exc->getMessage();
            $logger->error("%s: FAIL (%s)", $fn, $msg);
            $results[$fn]["success"] = FALSE;
            $results[$fn]["exception"] = $msg;
            $overall_success = FALSE;
            if (!$env['devmode']) {
              break;  // No point in continuing
            }
        }
    }
    $total_time = microtime(TRUE) - $start_ts;
    $resp = [
        "time" => $total_time,
        "success" => $overall_success,
    ];

    // ** Always print results
    $resp["checks"] = $results;
    header('Content-Type: text/json');
    echo json_encode($resp);
}

run_checks();

?>