508 Resource Limit Is Reached: when crawlers exhaust hosting
Why shared hosting shows 508 Resource Limit Is Reached, how to tell from Resource Usage and access logs whether crawlers caused it, and the fixes in order.
By Crawlito · 9 min read
In short
- On CloudLinux shared hosting, 508 Resource Limit Is Reached means the account hit its entry processes (EP) limit, the cap on concurrent PHP or CGI requests, which defaults to 20.
- Memory and process limits usually show up as 500 or 503 errors instead, so a 508 points specifically at too many dynamic requests running at once.
- Crawlers get the 508 too: Google slows crawling when it sees 5xx responses, ignores the content of the error page, and eventually drops URLs that keep returning server errors.
- Compare the times of EP faults in the Resource Usage page with your raw access logs to see whether a crawler burst lines up with the errors before changing anything.
- Fix in order: serve cached pages without PHP, close crawl traps, slow specific crawlers with documented controls, and raise limits only if real visitors still fill them.
On this page
- What "508 Resource Limit Is Reached" means
- What crawlers see when you hit the limit
- How to confirm crawlers are causing the 508
- Fix 1: serve cached pages without PHP
- Fix 2: close crawl traps
- Fix 3: slow specific crawlers with documented controls
- Fix 4: raise limits, last
- Frequently asked questions
- Check what crawlers get from your pages
"508 Resource Limit Is Reached" on shared hosting usually comes from CloudLinux: your account ran more PHP or CGI requests at the same time than its entry processes (EP) limit allows, so the web server refused the extra ones. Bursts of crawler traffic to uncached pages are a common cause, because each request holds an entry process for as long as PHP takes to build the page. Confirm it by matching the fault times in your hosting panel with your access logs, then remove the load with caching and crawl-trap fixes before you ask for higher limits.
This post covers the 508 itself: what the limit is, how to prove crawlers caused it, and what to fix on a shared plan. For the general method of measuring crawler cost per bot, see AI crawlers overloading your server, and for per-bot throttling rules, rate limit AI crawlers without locking out AI search.
What "508 Resource Limit Is Reached" means
Many shared hosts that run cPanel, Plesk or DirectAdmin use CloudLinux to put each account in its own Lightweight Virtual Environment (LVE) with fixed limits. CloudLinux's limits documentation lists them:
| Limit | What it caps | CloudLinux default | What happens at the limit |
|---|---|---|---|
| EP (entry processes) | Concurrent dynamic requests, such as PHP or CGI | 20 | Apache returns 508 |
| NPROC | Processes and threads in the account | 100 | Usually 500 or 503 |
| PMEM | Physical memory | 1024 MB | Usually 500 or 503 |
| SPEED | CPU, as a share of one core | 100% | Requests run slower |
| IO / IOPS | Disk throughput and operations | 1024 KB/s, 1024 ops/s | Requests run slower |
Your host may set different values; the defaults are only CloudLinux's starting point. The table shows why a 508 is a useful clue: CPU and disk limits slow you down, memory and process limits tend to produce 500 or 503, and only the EP limit produces 508. The Apache module mod_hostinglimits places each PHP or CGI request into your LVE. When the account already has as many entries as EP allows, it cannot place the new one and returns error 508. CloudLinux notes that LiteSpeed handles this limit differently, so on a LiteSpeed server the counts can differ from Apache's.
Two details matter for crawlers:
- EP counts requests in flight, not requests per minute. A crawler that asks for 10 uncached pages a second, each taking 2 seconds to render, keeps about 20 entries busy. That is the whole default allowance, and every visitor arriving at the same moment gets a 508.
- Static files do not use entries. mod_hostinglimits checks whether a request is for CGI or PHP, and CloudLinux says it does not limit static content by default. Images, CSS and HTML files that Apache serves directly do not pass through PHP, so they do not count against EP.
The standard meaning of 508 is unrelated. RFC 5842 defines it as "Loop Detected" for WebDAV operations. CloudLinux reuses the number for its own error page, so the status line alone does not tell a crawler what went wrong.
What crawlers see when you hit the limit
A crawler that arrives during an EP fault gets the same 508 a visitor does. It does not get your page.
Google documents how it handles server errors in its HTTP and network errors guide. It does not list 508 by name, but it covers the 5xx range:
- 5xx and 429 responses make Google's crawlers slow down temporarily, and the rate rises again gradually once the server returns 2xx.
- Google ignores the content of any 5xx response, so your error page text is not indexed.
- URLs already indexed are kept at first, but URLs that persistently return a server error are eventually dropped.
OpenAI, Anthropic and Perplexity do not publish how their crawlers treat a 5xx at the time of writing. The safe assumption is that any crawler receiving a 508 leaves without the content, including AI search crawlers such as OAI-SearchBot and PerplexityBot and user-triggered fetchers such as ChatGPT-User, which fetch a page because someone just asked about it. If Search Console shows server errors at the same time, server error (5xx) in Search Console explains that report.
How to confirm crawlers are causing the 508
Do not assume. A slow plugin, a cron job or a real traffic spike can fill entry processes just as well. You need two pieces of evidence for the same minutes: EP faults in the panel and a crawler burst in the logs.
Step 1: find the fault times in Resource Usage
CloudLinux's LVE Manager documentation describes the end-user Resource Usage plugin, which you open from the "CPU and concurrent connection usage" item in cPanel, Plesk or DirectAdmin:
- The Dashboard tab shows the account's general condition and any limits it hit.
- The Current usage tab has charts for entry processes and a Faults chart, plus a history table with average (A), limit (L) and faults (F) per period. Narrow the timeframe to find the exact periods with EP faults.
- The Snapshot tab shows process lists plus database and HTTP queries captured around a fault (database queries are cPanel only). The HTTP queries can show which URLs were being requested. CloudLinux notes that a snapshot is taken with a short delay, so it is close to, but not the same as, what was running at the moment of the fault.
If you cannot see these statistics, your host may have turned them off; CloudLinux gives administrators a "Hide LVE end user usage statistic" option. Ask support for the EP fault times instead.
Step 2: pull the access log for those periods
In cPanel, the Raw Access page downloads your access log as a .gz file; monthly archives are kept in /home/username/logs. The file name below is a placeholder, so use the name of the file you downloaded. Count the 508s per minute to confirm the fault windows from the server's side:
zcat example.com-ssl_log-Nov-2026.gz \
| awk '$9 == 508 {print substr($4, 2, 17)}' \
| sort | uniq -c | sort -rn | headThis assumes the combined log format, where field 9 is the status and field 4 the timestamp. Then see who was sending requests during the worst minute:
zcat example.com-ssl_log-Nov-2026.gz \
| grep '15/Nov/2026:03:14' \
| awk -F'"' '{print $6}' | sort | uniq -c | sort -rn | headIf one user agent accounts for most requests in that minute, and the same pattern repeats in other fault windows, you have your cause. Look at what it requested too: a crawler walking through ?filter=, ?sort= or internal search URLs is the usual picture on shops and directories.
Step 3: check the crawler is genuine
A user agent string is easy to fake. Before you write rules for "Googlebot" or "GPTBot", check the source IPs against the operator's published IP ranges or reverse DNS. Impostors are the easiest load to remove, and nothing you do in robots.txt will slow them.
Fix 1: serve cached pages without PHP
Because only dynamic requests use entry processes, the most effective fix is to answer crawler requests without starting PHP at all.
- Use full-page caching that writes static HTML. Some caching plugins and server modules serve stored HTML through rewrite rules or the web server itself, so a cached hit never enters PHP. Caches that still boot PHP on every request help less, because each request still holds an entry, only for a shorter time. Check your plugin's documentation for which mode it uses.
- Put a CDN in front for anonymous HTML. Requests answered at the edge never reach your account. Check that the cache key ignores tracking parameters, or each
?utm_variant becomes a fresh PHP request. - Make uncached pages faster. Shorter render time means each entry is released sooner. Halving page generation time roughly halves the entries a crawler holds at the same request rate.
The load reduction guide covers cache hit rates and conditional requests in more detail.
Fix 2: close crawl traps
Crawl traps are where crawlers spend the most entries for the least value: faceted filters, sort orders, calendars, internal search results and session IDs in URLs. These pages are rarely cached and are often the slowest to render.
Stop generating the links where you can, return 404 for invalid combinations, and disallow what must stay for people but has no search value:
User-agent: *
Disallow: /search
Disallow: /*?filter=
Disallow: /*?orderby=A crawler that follows a named group such as User-agent: GPTBot ignores the * group, so repeat these lines in every named group. Faceted navigation crawl traps covers filters in depth. Never let /robots.txt itself become a PHP request that can hit a 508: if your CMS generates it dynamically, a static file is safer on a tight plan. Google's robots.txt specification says that when robots.txt returns a 5xx error, Google stops crawling the site for the first 12 hours while it keeps retrying, then falls back to the last good copy for up to 30 days.
Fix 3: slow specific crawlers with documented controls
Once caching and traps are handled, slow any remaining heavy crawler with a method its operator documents, rather than blocking it.
- Googlebot: Google's crawl rate guide says to return 500, 503 or 429 instead of 200 for a short period, from a few hours to one or two days, with an optional
Retry-Afteron 503 or 429. It warns that URLs returning these codes for multiple days may be dropped from the index. If that is not practical, you can file a special request through the Googlebot report in Search Console, which Google says can take several days to evaluate. Google's robots.txt specification listscrawl-delayas unsupported. - Bingbot, ClaudeBot and CCBot: their operators document support for
Crawl-delayin robots.txt. The Crawl-delay guide lists who supports it. - Others: where an operator documents no rate control, a limit at the server or CDN is the remaining option, Blocking a training crawler in robots.txt is also an option: OpenAI says its GPTBot and OAI-SearchBot settings are independent, and Anthropic ties search visibility to Claude-SearchBot rather than ClaudeBot.
A deliberate 429 or 503 is better than a 508 for crawlers you want to keep: it is a documented signal to slow down, and you choose when it happens. On shared hosting you may not control the web server, but a CDN or your host's firewall may offer rate limiting. Some security tools on shared hosts block crawlers on their own; Imunify360 bot protection is worth checking if bots start getting 403s instead.
Fix 4: raise limits, last
If the logs show real visitors still filling entry processes after the fixes above, the plan is too small. Ask your host to raise EP and the related limits, or move to a plan with more. Raising limits first just moves the ceiling: an uncached crawl trap will fill 40 entries as easily as 20.
Frequently asked questions
Is 508 Resource Limit Is Reached caused by bots?
Sometimes. It means the account hit its entry processes limit, and crawler bursts to uncached pages are one common cause. Compare EP fault times with your access logs to know for sure.
Does a 508 error hurt SEO?
Short episodes mostly slow crawling. Google ignores the content of 5xx responses and eventually drops URLs that persistently return server errors, so repeated 508s on the same pages are a real risk.
Is 508 the same as Loop Detected?
The number is the same, but the meaning differs. RFC 5842 defines 508 as Loop Detected for WebDAV, while CloudLinux uses it to signal that the entry processes limit was reached.
What is a normal entry processes limit?
CloudLinux's default is 20 concurrent entries, but hosts set their own values. Your plan's limit appears in the L column of the Resource Usage page.
Will blocking AI crawlers fix a 508?
It can reduce load, but blocking search and user-triggered crawlers also stops those assistants from reading your pages. Cache, close traps and slow the heavy crawler first, and block only training crawlers or impostors if load remains.
Check what crawlers get from your pages
Once the 508s stop, make sure your fixes did not leave crawlers with a 403, a challenge page or an empty response. The free check fetches one page the way the main AI crawlers do, with no account needed.
Filed under CDN and bot protection. Topics: http-status, rate-limiting, caching, server-logs, troubleshooting