Skip to content

CloudLinux entry process (EP) limit reached on cPanel — when concurrency, not CPU, is the wall

An account throttles under load even though CPU looks fine. The wall is the EP limit — too many simultaneous requests. The fix is concurrency, caching, and finding what holds requests open.

When an account starts throttling — 508s, slow responses, requests queuing — the instinct is to blame CPU. But CloudLinux has a second wall that catches sites long before CPU does: the entry process limit, EP. EP counts how many dynamic requests an account may have running at the same time, regardless of how much CPU each one uses. A site can be sitting at 10% CPU and still hit its EP ceiling because too many requests are in flight concurrently — each one cheap, but too many at once. Hitting EP feels like a CPU problem (the site is slow under load) but the fix is completely different: you reduce concurrency, you do not buy more CPU. Telling them apart is the whole game.

What’s happening

EP (entry processes) is CloudLinux’s limit on simultaneous entries into the LVE — in practice, the number of PHP/CGI requests executing at once for that account. The default is often quite low (single digits to low double digits) because it is meant to be a per-account fair-share, not a throughput ceiling. When request number EP+1 arrives while EP requests are still running, CloudLinux either queues it or returns 508 to that visitor.

The key insight is that EP usage equals arrival rate × time each request stays open. A site hits EP not because requests are CPU-heavy but because they linger: a slow external API call inside a page, a wp-admin/admin-ajax.php heartbeat firing on every open tab, a long-running search query, a page that waits on a third-party script, or simply a burst of legitimate traffic. Each open request holds an entry process for its whole duration, so a page that takes 3 seconds ties up an EP slot 30× longer than a 100ms cached page. This is why caching is the single most effective EP fix: a cached page is served without entering PHP at all, so it consumes no entry process. The diagnosis is to confirm the faults are on EP specifically, then find what is holding requests open.

How to verify

Confirm the faults are EP, not CPU — they look identical from the browser but not in lveinfo:

# Faults by type for the account — look at the EP/EP-fault columns specifically
lveinfo --user=cpuser --period=1d --display-username

# Live: current entry processes vs the limit, across accounts
lvetop

# The account's EP ceiling
lvectl limits cpuser | grep -i ep

See how many requests are concurrently in flight and how long they run:

# Processes running in the cage right now (concurrent count = EP pressure)
ps -fu cpuser | grep -iE 'php|lsphp|cgi' | wc -l

# Slow / long-lived requests holding EP slots open
ps -eo pid,etimes,user,comm | awk '$3=="cpuser" && $2>5' | head

Find what kind of requests dominate — the EP cause is usually one hot endpoint:

# Hottest endpoints in the account access log (admin-ajax, search, an API path)
tail -500 /home/cpuser/access-logs/$(ls /home/cpuser/access-logs/ | head -1) \
  | awk '{print $7}' | sort | uniq -c | sort -rn | head

The fix

  1. Confirm it is EP, not CPU — lveinfo separates EP faults from CPU faults. If CPU is low but EP is faulting, do not raise CPU and do not chase slow queries first; the problem is concurrency. Every fix below targets concurrency, and applying CPU fixes to an EP wall wastes effort.
  2. Cache full pages so most requests never enter PHP — this is the highest-leverage EP fix because a cached response consumes zero entry processes. For a WordPress site, LiteSpeed Cache (on a LiteSpeed host) or a page cache serves the bulk of traffic from cache, leaving EP free for the requests that genuinely need PHP. A site that was hitting EP at 50 concurrent visitors often stops entirely once 90% of hits are cached.
  3. Shorten how long each request stays open — since EP usage is arrival rate × duration, halving page time halves EP pressure. Find slow pages (a page waiting on a slow external API, an uncached query, a third-party call) and make them fast or cache them. A request that finishes in 100ms instead of 3s frees its EP slot 30× sooner.
  4. Tame admin-ajax.php and the WordPress heartbeat — the heartbeat API fires repeated admin-ajax.php requests from every open admin tab, and a busy editor or a plugin abusing it can saturate EP on its own. Throttle or disable the heartbeat (the Heartbeat Control plugin, or limiting it to post-edit screens) so background polling stops eating entry processes.
  5. Block bot and crawler bursts that inflate concurrency — an aggressive crawler or a scraper hitting many dynamic URLs at once drives EP up without real users behind it. Identify it in the access log (step above), then block or rate-limit it at the firewall or via .htaccess. Bots hammering uncached pages are a common EP cause that no amount of limit-raising fixes.
  6. Move long jobs out of the request path — if pages are slow because they do heavy work inline (sending email, generating a report, calling an external service), that work holds an EP slot the whole time. Offload it to a background cron or a queue so the request returns immediately and releases its entry process. WordPress’s wp-cron run from system cron, rather than on page loads, is a concrete version of this.
  7. Raise the EP limit only for genuine concurrent load — after caching and trimming, if the account legitimately serves more concurrent dynamic requests than its EP allows (a real, sustained audience), raise the EP limit in WHM » CloudLinux LVE Manager for that account. Do this last, once you have confirmed the concurrency is real and not a cacheable or fixable pattern. Then re-check lveinfo and confirm EP faults drop to zero.

EP and CPU produce the same browser symptom and demand opposite fixes: CPU faults want less computation, EP faults want less concurrency — and caching, which removes requests from PHP entirely, is the lever that resolves most EP walls. Read the fault type first, then cache and shorten before you ever raise the limit. Stack Harbor diagnoses entry-process throttling under cPanel server management — start a conversation when an account hits its concurrency ceiling.